Flexibility options in electricity markets with high shares of renewable energies: An agent-based analysis of economic viability and system impacts

Flexibility options in electricity markets with high shares of renewable energies An agent-based analysis of economic viability and system impacts Dissertation zur Erlangung des Grades Doktor-Ingenieur der Fakultät für Maschinenbau der Ruhr-Universität Bochum von Felix Johann Nitsch aus Linz, Österreich Bochum, 2025 Eingereicht am 22. Juli 2025 Mündliche Prüfung am 22. September 2025 Erstgutachter: Prof. Dr. Valentin Bertsch Zweitgutachter: Prof. Dr. Dogan Keles “A system is more than the sum of its parts. It may exhibit adaptive, dynamic, goal-seeking, self-preserving, and sometimes evolutionary behavior.” — DONELLA H. MEADOWS, Thinking in Systems: A Primer Abstract As part of the energy transition, the electricity sector and its power generation infrastructure must be extensively transformed to reduce greenhouse gas emissions. Intermittent renewable energy sources are rapidly replacing fossil fuel-based electricity generation. Consequently, there is an increasing need for balancing supply and demand. Flexibility options are critical technological solutions for this challenge, however, significant knowledge gaps remain regarding their profitability and deployment in future electricity systems. This thesis contributes to closing these gaps by conducting agent-based electricity market simulations that target several key research objectives. First, it investigates operational strategies for flexibility options and their economic performance in electricity markets. The analysis reveals that strategies must consider the price impacts of flexibility options on both an individual and collective level. Second, the research examines profitable technical parameters and demonstrates that medium-term storage configurations likely offer the greatest revenue potential in day-ahead markets. This is particularly relevant in scenarios with increased competition between flexibility options. Third, the analysis identifies significant cannibalisation effects when substantial flexibility option capacity is integrated into the system. Due to their impact on electricity price dynamics, the overall revenue decreases, thus affecting profitability. These findings lead to the following recommendations for future research and policy. Endogenous modelling of large-scale flexibility options is necessary to understand future energy systems, as isolated analysis significantly underestimates market interactions. Additionally, the profitability of flexibility options is strongly linked to cost assumptions, particularly storage costs, which require consideration in investment decisions. Furthermore, planned investments in flexibility options should be accurately monitored, as cannibalisation effects and changing market dynamics will substantially impact profitability. The endogenous modelling of these complex phenomena represents a key methodological advancement of this thesis. This is achieved through extensions and application of a state-of-the-art electricity market model coupled with machine learning-based electricity price forecasting. This thesis contributes multiple software packages and data sets that adhere to FAIR principles, thereby enhancing reproducibility and facilitating future research. The presented work advances the understanding of the role of flexibility options in energy transition scenarios, while also revealing important avenues for future investigation. Key limitations to be addressed in further research include the focus on day-ahead market analysis rather than multi-market simulations, and simplified consideration of sector coupling dynamics. I Kurzzusammenfassung Im Rahmen der Energiewende müssen der Elektrizitätssektor und seine Stromerzeugungsinfrastruktur umfassend umgestaltet werden, um die Treibhausgasemissionen zu verringern. Erneuerbare Energiequellen ersetzen zunehmend die fossile Stromerzeugung. Durch die fluktuierende Einspeisung von Windund Solarenergie ist jedoch eine Ausbalancierung von Angebot und Nachfrage erforderlich. Flexibilitätsoptionen sind technologische Lösungen, um diese Herausforderung zu bewältigen. Dennoch bestehen erhebliche Wissenslücken bezüglich ihrer Rentabilität und ihres Einsatzes in zukünftigen Elektrizitätssystemen. Diese Dissertation trägt zur Schließung dieser Lücken bei, indem agentenbasierte Strommarktsimulationen durchgeführt werden, die mehrere wichtige Forschungsziele verfolgen. Zunächst werden Betriebsstrategien für Flexibilitätsoptionen analysiert und hinsichtlich ihrer Wirtschaftlichkeit auf Elektrizitätsmärkten bewertet. Es zeigt sich, dass Betriebsstrategien die Preisauswirkungen von Flexibilitätsoptionen sowohl auf individueller als auch auf kollektiver Ebene berücksichtigen müssen. Zweitens werden rentable Speichersystemparameter untersucht. Dabei werden mittelfristige Speicherkonfigurationen als Systeme mit dem besten Ertragspotenzial auf Day-Ahead-Märkten identifiziert. Dies gilt insbesondere in Szenarien mit verstärktem Flexibilitätswettbewerb. Drittens werden Kannibalisierungseffekte sichtbar, sobald erhebliche Flexibilitätskapazitäten in das System integriert werden. Aufgrund ihrer signifikanten Auswirkungen auf die Strompreisdynamik kommt es zu sinkenden Gesamterträgen, welche wiederum Auswirkungen auf die Rentabilität von Flexibilitätsoptionen haben. Auf Basis dieser Ergebnisse lassen sich die folgenden Empfehlungen für die künftige Forschung und Politik ableiten. Um künftige Energiesysteme besser zu verstehen, ist die endogene Modellierung von Flexibilitätsoptionen erforderlich, da bei einer isolierten Analyse die Marktinteraktionen erheblich unterschätzt werden. Die Rentabilität von Flexibilitätsoptionen hängt außerdem stark von Kostenannahmen ab, insbesondere von den Speicherkosten, die bei Investitionsentscheidungen sorgfältig berücksichtigt werden müssen. Geplante Investitionen in Flexibilitätsoptionen sollten darüber hinaus genau beobachtet werden, da Kannibalisierungseffekte und veränderte Marktdynamiken ihre Rentabilität erheblich beeinflussen werden. Die endogene Modellierung dieser komplexen Phänomene stellt einen wichtigen methodischen Fortschritt dieser Dissertation dar. Erreicht wird dies durch Erweiterungen und die Anwendung eines modernen Elektrizitätsmarktmodells, das mit Methoden des maschinellen Lernens für die Erstellung von Strompreisprognosen gekoppelt ist. Mit dieII F. Nitsch ser Dissertation werden mehrere Softwarepakete und Datensätze bereitgestellt, die den FAIR-Prinzipien entsprechen. Dadurch wird die Reproduzierbarkeit verbessert und weiterführende Forschung in diesem Bereich erleichtert. Zudem erweitert diese Dissertation das Verständnis der Rolle von Flexibilitätsoptionen in Energiewendeszenarien und zeigt gleichzeitig wichtige Aspekte für künftige Untersuchungen auf. Zu den wichtigsten Einschränkungen, die in der weiteren Forschung adressiert werden müssen, gehören die Beschränkung auf die Analyse des Day-Ahead-Marktes anstelle der Simulation mehrerer Märkte sowie die vereinfachte Berücksichtigung der Dynamik der Sektorenkopplung. III Declaration I hereby certify that I have independently authored the submitted dissertation titled Flexibility options in electricity markets with high shares of renewable energies:An agent-based analysis of economic viability and system impacts without external assistance, have not used sources other than those referenced in it, and have duly acknowledged all directly or indirectly adopted text passages, as well as utilised graphics, tables, and analysis programmes. Furthermore, I confirm that the electronic version presented corresponds to the written version of the dissertation, and that the thesis, in this or a similar form, has not been previously submitted and evaluated elsewhere. Erklärung Ich versichere an Eides statt, dass ich die eingereichte Dissertation mit dem Titel Flexibility options in electricity markets with high shares of renewable energies:An agent-based analysis of economic viability and system impacts selbstständig und ohne unzulässige fremde Hilfe verfasst, andere als die in ihr angegebene Literatur nicht benutzt und dass ich alle ganz oder annähernd übernommenen Textstellen sowie verwendete Grafiken, Tabellen und Auswertungsprogramme kenntlich gemacht habe. Außerdem versichere ich, dass die vorgelegte elektronische mit der schriftlichen Version der Dissertation übereinstimmt und die Abhandlung in dieser oder ähnlicher Form noch nicht anderweitig als Promotionsleistung vorgelegt und bewertet wurde. Felix Johann Nitsch Bochum, 21. Juli 2025 IV Acknowledgements First and foremost, I would like to express my sincere gratitude to my supervisor, Prof. Valentin Bertsch. I am deeply thankful for the constructive feedback, expert advice, engaging discussions, invaluable mentorship, and freedom he provided, all of which contributed greatly to my thesis and the enjoyment of my PhD journey. I would also like to express my heartfelt gratitude to my second examiner, Prof. Dogan Keles, for his thoughtful and thorough examination of my thesis. I sincerely acknowledge Prof. Patrick Jochem, Christoph Schillings, and all colleagues for providing the excellent conditions for this work at the Department of Energy Systems Analysis at the Institute of Networked Energy Systems of the German Aerospace Center (DLR). I also particularly thank Tatiana Borovleva and Melanie Knittel for their expert administrative support. Above all, I would like to express my sincere appreciation to Kristina Nienhaus, who initially encouraged me to start this thesis and supported me throughout the entire process. Her tireless dedication, positive attitude, and empathetic group leadership have been an essential foundation to enjoying my time at DLR. My heartfelt appreciation goes to Christoph Schimeczek, for his constant encouragement, the fruitful and stimulating discussions, and his excellent guidance in software development practices. I also extend my thanks to Ulrich Frey and Benjamin Fuchs, especially for proofreading early versions of my work. Additionally, special appreciation is owed to all colleagues and co-authors involved in any of the publications presented, A. Achraf El Ghazi, Evelyn Sperber, Florian Maurer, Hans Christian Gils, Johannes Kochems, Manuel Wetzel, Prof. Marc Deissenroth-Uhrig, Sebastian Wehrle, and Seyedfarzad Sarfarazi. Furthermore, I also would like to thank my PhD colleagues at the Ruhr-Universität Bochum, especially Jonas Finke, David Huckebrink, and Silke Johanndeiter for providing such a welcoming atmosphere. I was also privileged to spend time as a visiting researcher at the AIT Austrian Institute of Technology. The hospitality of Tara Esterl and her group, the Competence Unit Integrated Energy Systems (IES), made this stay very enjoyable and fruitful. V 1 Introduction This chapter provides an introduction to the content of this thesis, with Section 1.1 explaining the need for flexibility options (FOs) in future electricity markets. Section 1.2 provides a brief overview of the current state of the literature in this field, followed by a presentation of the identified research gaps and research targets. The structure of the thesis is outlined in Section 1.3. 1.1 Background and motivation The Paris Agreement, which was reached in December 2015, is a significant milestone defining global efforts to combat climate change (Schleussner et al. 2016). Its main objective is to limit the global temperature increase to well below 2°C above pre-industrial levels. It has shaped international climate policy and set ambitious targets for reducing greenhouse gas emissions. However, recent reports once again highlight the urgency of accelerating climate action (IPCC 2024). Therefore, a rapid and comprehensive energy transition from fossil fuels to renewable energy (RE) sources is critical. At the heart of the energy transition is the balance between three fundamental pillars: energy security, social and environmental considerations, and cost efficiency (Löschel et al. 2020). Energy security refers to the reliable availability of electricity at all times, which is essential for human welfare and economic stability. The increasing integration of RE sources, which are inherently variable and weather dependent, poses new challenges in maintaining this security. Therefore, it is necessary to ensure that electricity can be generated or stored to meet demand, even when RE sources are insufficient or unavailable at that time. Social and environmental considerations cover a wide range of factors, from public acceptance of RE technologies to biodiversity protection and land use concerns. Public perception of energy systems and their transformation can have a significant impact if there is opposition to certain infrastructure projects (Enserink et al. 2022). In addition, the environmental impact of the technologies used, whether through land use, resource extraction, or impact on ecosystems, must also be considered (Rahman, Farrok, and Haque 2022). Cost efficiency is another important aspect, as it is necessary to ensure that the transition to RE sources is economically viable. In liberalised markets, the energy transition depends on individual investments that are incentivised by attractive 1 F. Nitsch business cases. Investors must evaluate capital expenditures, ongoing operational costs, and expected revenues within a low-emission energy system. At the same time, policy and market frameworks should also account for the broader, long-term economic benefits of achieving climate targets and mitigating climate change at the system level (Bogdanov, Ram, et al. 2021; Osman et al. 2023). Therefore, it is critical to consider both these microand macroeconomic perspectives for the success of the energy transition. The ultimate goal of the energy transition is to reduce global greenhouse gas emissions (Rogelj et al. 2019). This thesis thereby focuses particularly on the electricity sector, which is a major contributor (IEA 2024). To achieve the emission reduction targets, new power plants must be implemented, especially those based on RE sources, such as wind, solar, and hydropower (Twidell 2021). It also involves electrification of various sectors, which will increase demand for electricity (Staffell and Pfenninger 2018). In particular, the transition to electric vehicles, the electrification of heating systems through heat pumps, and the transformation of industrial processes away from fossil fuels all contribute to this increase in demand. Growing shares of fluctuating RE sources also require a flexible and responsive energy system that can adapt to changes in supply and demand across timescales from seconds to seasons (Leonard, E. Michaelides, and D. Michaelides 2020). This flexibility can be provided through various technological solutions, collectively referred to as flexibility options (FOs) (Zöphel et al. 2018). FOs can be characterised by several key parameters such as capacity (the total amount of energy that can be stored, i.e., MWh), power (the rate at which energy can be charged or discharged, i.e., MW), ramping capability (the rate of change of power, i.e., MW/min), and cost (both capital and operating costs, i.e., EUR/MW, EUR/MWh) (Alizadeh et al. 2016). A well-established example of a FO is pumped hydro storage (Babatunde, Munda, and Hamam 2020). This mature and widely used technology stores energy by pumping water uphill to a reservoir during periods of low demand/price and releasing it during periods of high demand/price to generate electricity, profiting from this arbitrage. Although it offers significant capacity, its scalability can be limited by geographical restrictions that require significant up-front investment (Hunt et al. 2020). In contrast, battery storage systems, particularly lithium-ion batteries, have experienced rapid cost reductions and efficiency improvements in recent years (Cole, Frazier, and Augustine 2021). These systems are highly scalable and can provide short-term flexibility services, making them an increasingly attractive solution for integrating RE sources into the energy system. Another concept for providing flexibility is “sector coupling”, which refers to the integration and coordination of different energy sectors (Fridgen et al. 2020). This involves linking the electricity sector with other areas, such as heating, industry, and transportation (Orths 1 Introduction 2 F. Nitsch et al. 2019). Although, it is estimated that the flexibility potential is significant (Bernath, Deac, and Sensfuß 2021), it is challenging to assess specific economic feasibility as all solutions compete with each other (Ramsebner et al. 2021). As with RE technologies, there are additional considerations in addition to technical specifications that are relevant to the successful application of FOs. These are related to scalability and expansion potential. Social acceptance can have a significant impact on the feasibility of projects, as shown by public opposition to new pumped hydro projects in environmentally sensitive or scenic areas (Pickard 2011). Many technologies are still evolving and may not yet be available on a scale or at the cost required for widespread deployment. This may already affect today’s investment decisions (Koltsaklis and Knápek 2023). As these projects typically are large in capital with long payback periods, an accurate evaluation of their economic potential is essential (Keles 2013). Therefore, a thorough ex ante analysis is needed to assess the economic and operational viability of these FOs for application in future market scenarios (Ölmez, Ari, and Tuzkaya 2024). Several analyses have already been carried out in this area and will be discussed in the next section. 1.2 Research targets Existing studies of FOs in electricity systems generally fall into two categories. The first category focuses on individual devices, examining the technical and economic performance of individual storage devices without considering their market implications (Elalfy et al. 2024). While these studies provide valuable information on operational characteristics, they often use historical price time series. This is a critical limitation because they cannot capture the potential impact of large-scale FOs deployment on electricity market dynamics and profitability (Lund et al. 2015). The second category uses system optimisation with a “central planner” approach, assuming perfect coordination between all components of the system to achieve a global optimum (Mancò et al. 2024; A. M. Barbosa et al. 2024). These models often assume perfect foresight, which does not reflect the reality of decisionmaking under multiple uncertainties (Fodstad et al. 2022). In addition, they tend to be computationally expensive, which limits the number of scenarios that can be explored (Ma and Nakamori 2009). Furthermore, energy systems are highly complex because they are interconnected across multiple dimensions, from technical and social to environmental concerns (Bale, Varga, and Foxon 2015). Therefore, models studying FOs must capture not only technical aspects, but also social aspects such as the behavior of individual actors (Pfenninger 2024). These limitations and challenges highlight the need for modeling 1 Introduction 3 F. Nitsch approaches that can handle such complexity. What is missing is a comprehensive analysis that accounts for agent behaviour in electricity markets characterised by increasingly high RE shares while explicitly considering operational uncertainty and forecasting limitations (Bessa et al. 2019). Future electricity markets will exhibit fundamentally different price dynamics. On the one hand, RE will increase the price spreads between peak (low RE generation) and off-peak prices (high RE generation). On the other hand, various FOs will exploit these attractive spreads through arbitrage, thereby reducing them over time. Yet current FO profitability assessments remain largely based on historical market patterns that neglect these dynamic interactions. Market imperfections and uncertainty, such as participants not bidding at marginal cost, can further significantly affect revenue potential, but are often overlooked in current studies (Reza et al. 2023). Moreover, the impact of strategic individual behaviour by market participants, especially FOs, should be accounted for in these transformed market conditions (Siala et al. 2022). To address these complex interactions between individual agent decisions and system-wide market outcomes, electricity market simulation methods offer a proven analytical framework (Weidlich and Veit 2008). Agent-based models (ABMs) are a particularly powerful approach within this framework, as they place individual actors at the centre of analysis and examines how system-level effects emerge from their strategic interactions (J. Castro et al. 2020). Despite this potential, more research applying ABMs to energy systems analysis (ESA) is needed to explicitly address these identified research gaps (Heider et al. 2021). There is also a gap in open research software and data availability. Although there are notable developments in the domain of open science in ESA (Gils et al. 2022), open software that is well documented, thoroughly tested, and modularised for flexible application can further contribute to the field. Adherence to the findable, accessible, interoperable, reusable (FAIR) principles (Barker et al. 2022) for research software would enhance the transparency and reproducibility of the results. Open data repositories, which can be freely used, facilitate simple and convenient application and extension of ESA models, and thus accelerating progress in the field (Chang et al. 2021). This thesis addresses the identified research gaps through advancements in two interconnected domains: energy economics and energy informatics. Within the energy economics domain, the following research targets are identified. 1.1 Identify operational strategies for FOs that perform reliably in future electricity market scenarios with increasingly high RE shares. 1.2 Evaluate how technical specifications (e.g., capacity & power) of FOs influence refinancing potential under varying cost assumptions. 1 Introduction 4 F. Nitsch 1.3 Quantify the impact of increasing market penetration of competing FOs on their profitability in day-ahead market (DAM). From an energy informatics perspective, the objective is to extend and enhance existing models to enable a detailed representation of FOs and their market interactions, as outlined below. 2.1 Expand ABM to simultaneously capture both individual FO economics and their collective impact on system dynamics. 2.2 Develop modular open-source software packages to enhance reproducibility and facilitate comparative assessment of FOs. The novelty of this research lies in the combination of different modelling techniques, from ABM to optimisation and machine learning (ML), to provide a more comprehensive understanding of FO in systems with high shares of RE. Given that individual actions can aggregate into emergent, often unintended outcomes at scale (Schelling 2006), I explicitly examine this phenomenon in the context of increasing FO deployment. Furthermore, this work contributes numerous software packages to the field, adhering to open science practices, specifically following the FAIR principles (Barker et al. 2022), thus facilitating future research. By bridging the gap between theoretical optimality and realisable outcomes, this research provides valuable insights for policy makers, investors, and system operators navigating the complexities of the energy transition. Figure 1.1: Spatial, temporal, technological, and economic scopes of the applied energy system models. Illustration based on Cao et al. 2021. 1 Introduction 5 F. Nitsch Figure 1.1, which is based on an illustration by Cao et al. 2021, shows the spatial, temporal, technological, and economic scope of the ESA models applied in this thesis. The spatial scope primarily targets the German DAM zone with extensions to neighbouring market zones. Temporarily, the analysis ranges from accounting for RE fluctuations to seasonal weather effects, partially covering investment planning perspectives. Although the work features some aspects of sector coupling and system services, the technological scope focuses primarily on the mix of power generation. From an economic perspective, I address the interactions between companies who could operate a FO, and whole markets. 1.3 Structure of the thesis The structure of the thesis is presented in Figure 1.2. Chapter 2 provides a high-level presentation of the applied materials and methods. This includes a detailed description of the applied ESA models in Section 2.1, complementary tools and methods in Section 2.2, and model calibration in Section 2.3. The four main research papers are presented in Chapter 3, specifically in Sections 3.1 to 3.4, each with bibliographic information including an executive summary and author contributions. A synthesis of all the research papers presented is provided in Section 3.5, which describes how the papers contribute to filling the identified research gaps. Chapter 4 outlines the limitations of the presented theses, summarises the overarching findings, and draws conclusions. The Appendix contains supplementary peer-reviewed and supplementary non-peer-reviewed papers in Chapters A and B respectively. A short academic curriculum vitae is presented in Chapter C. Figure 1.2: Outline of the thesis. 1 Introduction 6 2 Material and methods This chapter presents an overview of the approaches and models used to address the research targets presented. In contrast to the individual materials and methods sections in the research papers (see Chapter 3), this chapter provides an overview of the primary methods used and lists my major methodological contributions. In the beginning, the field of energy systems analysis (ESA) is introduced in Section 2.1 by outlining two prominent and widely used approaches, agent-based models (ABMs) (Subsection 2.1.1)and energy systems optimisation models (ESOMs) (Subsection 2.1.2). An overview of complementary methods, such as machine learning (ML) and model coupling, is given in Section 2.2. Section 2.3 on scenario definition and data provides insights in the modelling assumptions, model validation, and also tools which are used to perform multi-scenario analysis. 2.1 Energy system models System models in the energy domain were already in use in the mid 20th century (Barnett 1950). During the oil crisis in the 1970s, ESA contributed to the public discussion (Meadows et al. 1972; Hoffman and Wood 1976). More recently, the transition to low-carbon energy systems and its implications along the transformation lead to increasingly complex problems (Nakata, Silva, and Rodionov 2011). It is important to find an optimal allocation of power plant technologies under the consideration of meeting the demand while staying below the emission limits (Davis et al. 2018). For many stakeholders, from policy makers to investors, it is crucial to know how target systems may be specified. This allows them to pursue relevant investments and avoid potential lock-in effects (Bertram et al. 2021). ESA modellers have faced new challenges in the last decade due to the increasing complexity of both real-world systems and the models employed (Pfenninger, Hawkes, and Keirstead 2014). There is a wide range of ESA models available, each with its own temporal, technological, sectoral, or spatial focus (Fodstad et al. 2022). Increasing computational capacities foster the development and application of comprehensive ESA models, realising detailed model coupling workflows. Open modelling approaches, including methodological transparency and data accessibility, enhance the credibility and impact of energy systems research (Pfenninger, Hirth, et al. 2018). 7 F. Nitsch In this thesis, I focus on two main approaches, ESOMs and ABMs. ESOMs are widely applied to identify optimal energy systems under the consideration of, e.g., emission constraints (Hoffmann et al. 2024). They can also be used for dispatch planning and unit commitment purposes. ABMs focus on individual or prototypical agents and model the outcomes by agent interaction. These models can be used to assess specific system designs in an exploratory way, combining microand macroeconomic viewpoints. These analyses may be attributed to the field of energy economics, which studies the supply and demand of energy. In this regard, electricity markets are a fundamental component of energy systems, as they facilitate the allocation of energy from producers to consumers. As the energy transition leads to more electrification across all sectors, the demand for electricity is expected to increase (IEA 2024). Electricity markets will therefore become more important. In the context of this research, my focus is primarily on day-ahead markets (DAMs), which provide crucial price signals to all market actors in several markets (SilvaRodriguez et al. 2022). In particular, the electricity exchange collects bids and asks, and periodically clears the market. The electricity market is shaped by various actors, including producers, consumers, traders, flexibility providers, and regulators. These actors can be represented as (prototypical) agents in an ABM. 2.1.1 Agent-based simulation ABM is a computer simulation approach explicitly modelling interactions between agents that can reveal emergent behaviour (Helbing and Balietti 2012). Starting from an initial scenario configuration, ABM allows exploratory analysis of potential outcomes (Tesfatsion 2006). In the context of ESA, ABM offers a number of significant strengths (Ringler, Keles, and Fichtner 2016). First, by incorporating the perspective of individual actors, researchers can identify potential emergent effects (Frey, Klein, et al. 2020). Secondly, the use of heterogeneous agents in an ABM simulation allows for the representation of a range of actor characteristics, including their objectives, risk profiles, information levels, and interactions with their environment (Kraan, Kramer, and Nikolic 2018). Thirdly, ABM offers superior practical applicability to address real-world energy transition challenges while maintaining computational feasibility (Hansen, Liu, and Gregory M. Morrison 2019b). Fourthly, uncertainties of various origins can be considered, ranging from system design to imperfect information (Hansen, Liu, and Gregory M Morrison 2019a). Finally, in contrast to optimisation models, there is no global objective function to be maximised or minimised (Ma and Nakamori 2009), however, individual agents can use optimisation models for their own decision-making (Klein, Frey, and Reeg 2019). 2 Material and methods 8 F. Nitsch It is also important to consider the challenges of ABM, which can be caused by parameterisation issues (Hammond 2015). As (prototypical) actors are the core of any ABM problem, detailed knowledge on the individual actor’s is critical (Janssen and Ostrom 2006). Due to its exploratory nature, even minor changes in input parameter configurations can yield significantly different results (J. Castro et al. 2020). Therefore, model validation, such as backtesting against historical data, is an important step in ABM application (F. Maurer et al. 2024). 2.1.1.1 AMIRIS The open Agent-based Market model for the Investigation of Renewable and Integrated energy Systems AMIRIS is a state-of-the-art ABM to analyse renewable energy (RE) market integration and policy effects (Schimeczek, Nienhaus, et al. 2023). It was created by the German Aerospace Center (DLR) over a decade ago and has been continuously developed ever since, with the software being openly available since 2021. The various types of agents perform tasks based on their own inputs and their surrounding environment, as illustrated in Figure 2.1. Modellers can adjust the level of information for each agent individually by defining the so-called contracts. They represent a formalised method of exchanging information at a specified time between agents. Figure 2.1: AMIRIS model structure revealing agent types and their connection (Schimeczek, Nienhaus, et al. 2023). 2 Material and methods 9 F. Nitsch The focus of AMIRIS is the DAM, where uniform electricity prices are determined in an hourly resolution. In order to achieve this, the energy exchange agent collects bids and asks from the relevant agents, either providing electricity supply or requesting to meet electricity demand. Electricity supply can be provided by conventional power plants or a RE plants. These units can be as detailed as a single block of a power plant to an aggregated fleet of power plants. Each power plant operator maintains a connection with a respective trader whose objective is to sell their power generation potential on the DAM. In order to facilitate this, the trading agent receives marginal cost information from the power plant operators, which are determined by operational costs, fuel costs, and emission certificate costs. Traders are then able to add optional markups to marginal costs, accounting for ramp-up costs (Liberopoulos and Andrianesis 2016), and markdowns, accounting for ramp-down costs (Pape, Hagemann, and Weber 2016), both of which represent non-convex nature of such costs (Makkonen and Lahdelma 2006). Demand traders are defined as actors who seek to procure a specified quantity of energy at a designated point in time. A “value of lost load” is assigned to the bid, which represents the maximum price the bidder is willing to pay for the requested demand. Flexibility option (FO) agents, such as storage operators, control a storage device, enabling them to purchase and sell energy to the market according to their internal state of charge (SoC). As this is a time-dependent matter, the operator must develop an operational schedule based on a forecasted electricity price. This optimisation problem is solved using dynamic programming (Bellman 1957) and discrete levels of SoC. From a strategic point of view, the objective may be to maximise profits or to minimise overall system costs. A policy agent enables the examination of various policy instruments with the objective of supporting the operations of RE power plants. Furthermore, additional neighbouring DAMs can be interconnected through a central market coupling agent (Nitsch and El Ghazi 2023). Time-dependent transmission capacities between market zones account for the comprehensive states of interconnected electricity markets. The central coupling algorithm is designed with the objective of optimising the collective welfare of all interconnected markets. A comprehensive account of all agent types can be found in the AMIRIS-Wiki1. The Python package AMIRIS-Py provides a convenient wrapper for all modelling tasks (Schimeczek and Nitsch 2024). Specifically, AMIRIS-Py handles the installation, setup, model execution, and post-processing of the results. With AMIRIS-Examples, there is also an open scenario data collection available (Nienhaus et al. 2025). This ensures an easy on-boarding for new users, but also convenient workflow adaptations for more ex1https://gitlab.com/dlr-ve/esy/amiris/amiris/-/wikis 2 Material and methods 10 F. Nitsch (a) Training workflow in focapy. (b) Evaluation workflow in focapy. Figure 2.5: focapy class diagrams. 2.2.1.2 AMIRIS-PriceForecast FOs, such as battery storage systems, play a key role in most energy scenarios by shifting energy over time. Both in reality and in many electricity market simulations, FOs typically require electricity price forecasts to optimise their operational schedules. For instance, Figure 2.6 shows how this procedure is implemented in AMIRIS. Specifically, aForecastAgent collects preliminary bids and asks from supply and demand traders, except those of the FOs. It then sends a “naive” - as it does not include any actions of the FOs - electricity price forecast to the FOs. This electricity price information, either a point in time or a merit order information, is used to optimise the bidding behaviour of the FOs. If there is only a single simulated FO in a scenario, it can accurately consider its own impact on the electricity price forecast, and therefore it can be considered a perfect forecasting situation. However, when multiple FOs are simulated simultaneously, the naive electricity price forecast does not take into account the bidding behaviour of (competing) FOs. Therefore, the revenue of individual FOs are negatively affected by the so-called “avalanche effect”, where multiple actors react on the same signal, thus they may not achieve their expected outcome (Kühnbach, Stute, and Klingler 2021). This can also occur when the collective impact of individual actions reduces the overall revenue potential through “cannibalisation effects,” as observed for RE sources (Hirth 2013) or FOs (Ölmez, Ari, and Tuzkaya 2024). 2 Material and methods 17 F. Nitsch Figure 2.6: Flexibility option implementation in AMIRIS. To address the limitations of these naive forecasts, I have developed a dedicated Python package that extends the current forecasting procedure in AMIRIS by coupling external forecasting algorithms (Nitsch and Schimeczek 2025b). The package AMIRISPriceForecast (Nitsch and Schimeczek 2025a) hosts modular forecasting architectures from simple time-shift models (Hyndman 2018) to comprehensive NN based forecasts like Transformers (Lim, Arık, et al. 2021). Figure 2.7 shows the interaction between PriceForecasterApi, a new Java agent designed to improve electricity price forecasting, its ForecastClient, such as a StorageTrader in AMIRIS, and the external AMIRIS-PriceForecast model itself (Nitsch 2025b). The ForecastClient requests an electricity price forecast for a defined period, e.g., the next 24 hours in the simulation. The PriceForecasterApi agent receives the request and passes it to AMIRIS-PriceForecast where the actual forecast is calculated. Once the forecast electricity price time series has been received, the PriceForecasterApi agent returns the price time series to its client. Communication between AMIRIS and AMIRISPriceForecast is handled by HTTP requests with data in JSON format using FastAPI2. AMIRIS waits for the response message from the external model and then continues the simulation. 2https://github.com/fastapi/fastapi 2 Material and methods 18 F. Nitsch Figure 2.7: Enhanced electricity price forecasting in AMIRIS by calling an external forecasting model. During a stress test (Nitsch, Sperber, et al. 2025), AMIRIS-PriceForecast was called every simulation hour with a payload consisting of 16 time series, each containing 168 time steps. The communication overhead between the models had a minimal impact on the overall runtime. The main cost driver is the execution of the ML prediction itself. In general, a yearly AMIRIS simulation with hourly resolution takes about 20 seconds when using the external model, compared to just 10 seconds for a standard run. If runtime of the model is critical, users can reduce the number of calls to AMIRISPriceForecast by specifying an extension window that contains additional forecasted time steps. If the realised electricity prices remain below the specified tolerance threshold, PriceForecasterApi will fulfil the forecast requests using its internal memory. Table 2.1 provides an overview of the forecast algorithms available in AMIRISPriceForecast and accessible from AMIRIS. At the time of writing, there are three “naive” models and two NN models. The first group are TimeShift models, which use the last 1, 24 or 168 hours as a forecast and shift it into the future. If the requested forecast horizon exceeds the last prices, they are propagated continuously. For example, 2 Material and methods 19 F. Nitsch when pTrepresents the price at time T,ˆpT+h|Tdenotes the forecasted price at time T+h which follows a 24 hour pattern: ˆpT+h|T=pT+h−24 There is also a SimpleNN model which consists of PyTorch (Paszke et al. 2019) layers. The exact specification and depth can be parameterised using a flexible YAML configuration file. Finally, there is a Transformer (Lim, Arık, et al. 2021) implementation based on the darts (Herzen et al. 2022) package. This architecture has proven to be a powerful yet flexible algorithm for modeling future electricity markets (Nitsch, Schimeczek, and Bertsch 2024). To account for forecast uncertainty through ensemble forecasting, a common approach in practice, the model can be called multiple times to generate individual forecasts. This then enables the application of probabilistic operational strategies. As the field of time series forecasting advances rapidly, AMIRISPriceForecast was designed in such a way that adding new forecasting architectures, such as TabPFN-TS (Hoo et al. 2025) is straightforward. Depending on the complexity of the underlying energy system scenario, time series forecasting can be a challenging task that demands detailed domain knowledge for selecting appropriate models and (hyper) parameters (Keles et al. 2016). Within this thesis, Paper III investigates anticipated shifts in energy transition scenarios, focusing specifically on how electricity price dynamics change in systems with significant shares of RE. Table 2.1: Available forecast algorithms in AMIRIS-PriceForecast. Name Description Probabilistic Features TimeShift1 Predictor using last hour as forecast – Past Targets TimeShift24 Predictor using last 24 hours as forecast – Past Targets TimeShift168 Predictor using last 168 hours as forecast – Past Targets SimpleNN Basic NN based on PyTorch (Paszke et al. 2019) – Past & Future Covariates Transformer Temporal Fusion Transformer (Lim, Arık, et al. 2021) Yes Past & Future Covariates Figure 2.8 shows the workflow used for automated ML training. In contrast to focapy, this workflow extends the functionality by also generating synthetic training data using AMIRIS-Scengen (see Section 2.2.3). Specifically, user-generated scenario templates are converted into actual AMIRIS simulation input files and prepared for training. Then a ML of choice is trained within the AMIRIS-PriceForecast package. Finally, both model and the scenario data are saved to disk. 2 Material and methods 20 F. Nitsch Figure 2.8: AMIRIS-PriceForecast training workflow. After model training is completed, it can be used for inference during an AMIRIS simulation. The associated workflow, shown in Figure 2.9, handles AMIRIS-PriceForecast and the actual AMIRIS simulation. First, the trained model is loaded and the forecasting service endpoint is set up. Second, the AMIRIS simulation is started, where the PriceForecasterApi agent can request external electricity price forecasts. Once the model has run, all programmes, including the forecasting service endpoint, are shut down. Figure 2.9: AMIRIS-PriceForecast inference workflow. 2.2.2 Model coupling and workflow management Models in ESA are often specialised in a particular aspect or focus on a method. In order to extend the capability and functionality of a model, we can use model coupling. Model coupling can be achieved at different levels, ranging from loose/soft types (e.g. data exchange) to tight/hard types (e.g. interconnected modules) (Yourdon and Constantine 1979). In this thesis, there are different implementations of model coupling, which are outlined below. 2 Material and methods 21 F. Nitsch In the first paper (Nitsch, Deissenroth-Uhrig, et al. 2021) the ABM AMIRIS is coupled with a dispatch optimisation model from the perspective of a storage operator trading on the DAM and automatic Frequency Restoration Reserves (aFRR). In the second paper (Nitsch, Wetzel, et al. 2024) a coupling between two comprehensive ESA models, namely REMix (Wetzel et al. 2024) and AMIRIS, is implemented. For this, REMix defines the power plant park from an optimal system perspective, while AMIRIS analyses the performance of the Carnot battery storage. The technical component of such a coupling is ideally performed by a workflow manager. In this thesis, the Python workflow manager ioproc (Fuchs et al. 2020) is used for such tasks. It provides a powerful framework for flexible coupled workflows, such as iog2x (Nitsch, Schimeczek, Wetzel, et al. 2023) which is also used in (Nitsch, Wetzel, et al. 2024). 2.2.3 Multi scenario analysis using AMIRIS-Scengen Conducting multiple model runs enables comprehensive analysis of the scenario dynamics and parameter sensitivity. However, this requires variation of the input data, which can be tedious and error-prone when created manually. I have therefore developed the scenario generator AMIRIS-Scengen (Nitsch, Frey, and Schimeczek 2023). AMIRIS-Scengen is a lightweight Python package that runs with minimal requirements on Windows and UNIX-based operating systems. It is designed for the electricity market model AMIRIS (Section 2.1.1.1), but with little modification it could also be applied to other FAME models (Section 2.1.1.2). As shown in Figure 2.10, AMIRIS-Scengen divides the workflow into dedicated tasks, facilitating integration with external programmes and enabling model coupling (Section 2.2.2). The core functionality of AMIRIS-Scengen is to randomly generate AMIRIS scenarios based on a user-defined set of rules, assess the plausibility of the scenarios, feed the scenarios to AMIRIS where they are executed, and evaluate the final results. Specifically, the user defines a scenario that allows parameters to be associated with certain keywords, such as •a random draw of discrete values of type string, integer or float, •a random file path within a given directory, •a random integer or float between a minimum and maximum value, or •a fixed value. In addition, agents and their associated contracts can be automatically added. Finally, the user specifies the number of different scenarios to be generated. Reproducibility 2 Material and methods 22 F. Nitsch is ensured by a trace file that stores the random seed and a counter for the number of scenarios generated, allowing AMIRIS-Scengen to assign a unique and traceable identifier to each generated scenario. The generated scenario is passed to a pre-simulation check (“Estimation” stage) where the user can define a set of rules that make certain parameter configurations invalid. If this is the case, the current scenario is discarded and a new scenario is generated. The valid scenarios are then passed to AMIRIS, which performs the actual simulation. The processing of the result files is carried out in a postsimulation check (“Evaluation” stage) where the user can define certain conditions for a scenario, e.g. uncovered load due to missing electricity generation potential. Similarly to the “Estimation” stage, scenarios that fail any of the tests are discarded. This workflow ensures that the user can easily and reliably create numerous AMIRIS scenarios in a flexible and convenient way, see also Figure 2.11. AMIRIS-Scengen is employed in this thesis to generate training, test, and validation data sets for the paper on time series forecasting (Nitsch, Schimeczek, and Bertsch 2024) and to conduct multi-scenario analysis in preparatory work (Nitsch and Schimeczek 2024) for main Paper IV (section 3.4), which assesses FOs potential (Nitsch, Schimeczek, and Bertsch 2025). Figure 2.10: AMIRIS-Scengen model workflow showing the different modules and their interactions (Nitsch, Frey, and Schimeczek 2023). 2 Material and methods 23 F. Nitsch Figure 2.11: Multi-scenario analysis using AMIRIS-Scengen. 2.3 Model calibration Model calibration and back-testing are critical steps in the development of reliable analytical tools for ESA. The accuracy and impact of any modelling approach depend significantly on both the quality of the input data and the structural validity of the model itself. Comparison with historical DAM results ensures that models can reproduce historical patterns before being applied to future scenarios or policy assessments. 2.3.1 Data This thesis uses open data sources primarily, adhering to the principles of transparency and reproducibility of research. The AMIRIS-Examples data set (Nienhaus et al. 2025), first published in 2022 and continuously updated since then, serves as the primary source of AMIRIS data, providing comprehensive data under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence. This data set integrates information from several sources, including the European Network of Transmission System Operators for Electricity (ENTSO-e), the German Electricity Market Data Platform (SMARD), the Federal Statistical Office of Germany (Destatis), the Federal Ministry for Economic Affairs and Climate Action (BMWK), the European Power Exchange (EPEX SPOT), and the Austrian Power Grid (APG). The compilation includes key parameters for energy 2 Material and methods 24 F. Nitsch system modelling, including generation profiles across different technologies, temporal demand patterns, installed power plant capacity, and costs. At the time of writing, the data mainly covers the German and Austrian market zones and spans historical years from 2015 to 2019. The ongoing expansion efforts aim to include additional years and market zones, which will further enhance the utility of the data set for a broader analysis and long-term trend assessment (Nitsch, Schimeczek, Nienhaus, et al. 2025). 2.3.2 Benchmarking Benchmarking against other established models is an effective way to understand the specifics of a model and identify strengths and limitations, while promoting transparency within the research community. The AMIRIS model was systematically compared with the ASSUME market model (F. Maurer et al. 2024). This comparison showed that both models can reproduce the dynamics on the German DAM with errors below 6.4 EUR/MWh. The dispatch of power plants was also compared, where AMIRIS achieved a good fit of the simulation results to the historical data from 2019. Additional AMIRIS back-testing against historical data was performed for the German DAM (Nitsch, Deissenroth-Uhrig, et al. 2021) and the Austrian DAM (Nitsch, Schimeczek, and Wehrle 2021). These validation analyses examine how accurately the model reproduces historical market clearing prices and dispatch patterns when calibrated with historical input data. Both studies identified specific areas for model refinement, in particular regarding the representation of strategic bidding behaviour of FOs and RE operators in situations where negative prices occurred in reality. 2 Material and methods 25 3 Publications This chapter provides a summary of the publications included in this cumulative thesis. The thesis comprises four research papers, three of which have been published in peer-reviewed journals and one of which is currently under review. For each full paper, Sections 3.1 to 3.4 provide bibliographic information. This includes an executive summary, which differs from the individual abstract by providing targeted results without repeating extensive general contextualisation. I then describe the contributions of myself and all co-authors. These descriptions are in line with the contribution statements included in the publications, but are more detailed in certain aspects. In addition, the journal articles are included in full text in their respective journal format1. Finally, in Section 3.5, I elaborate on how the publications are connected and form a coherent overall structure, contributing to the individual research targets identified in Section 1.2. For this, some additional complementary papers are provided in the Appendix A (peer-reviewed) and Appendix B (non-peer-reviewed). 1Abbreviations and reference numbering are specific to each individual article. Additionally, the journal articles use American English while this thesis uses British English. 26 Applied Energy 298 (2021) 117267 5 using the CPLEX solver [61]. It is assumed that the BSS is prequalified for trading on the DA market and aFRR market aiming to maximize its total revenue under perfect foresight over the observation period of one year. The function RevenueTotal =∑ 8760 i=1 RevenueDA,i+RevenueaFRR,i(5) describes the total revenues consisting of the summed revenues in hour i in the two markets which the storage operator tries to maximize. The revenues from trading on the DA market are defined by RevenueDA,i=pDA,i*(EnergyDA Sell,i−EnergyDA Buy,i)(6) with pDA,i as the price at the DA market, EnergyDA Sell,i as the energy sold at the DA market and EnergyDA Buy,i as the energy bought at the DA market in hour i. The revenues from the aFRR consist, on the one hand, of the income from the provision of power RevenueaFRR,Power,i in positive or negative direction RevenueaFRR,Power,i=paFRR,Power,positive,i*PoweraFRR,positive,i +paFRR,Power,negative,i*PoweraFRR,negative,i(7) which are awarded with the prices paFRR,Power,positive,i and paFRR,Power,negative,i assuming that the BSS places its bids at the same price as the most expensive power plant which is still in the market; on the other hand, the revenue ReveueaFRR,Energy,i from the actual energy flows ReveueaFRR,Energy,i=paFRR,Energy,positive,i*EnergyaFRR,positive,i +paFRR,Energy,negative,i*EnergyaFRR,negative,i(8) which is awarded with the prices paFRR,Energy,positive,i and paFRR,Energy,negative,i when reserve capacities are actually called.By its specifications, the BSS is equipped with technical parameters that characterize its performance. First of all, the state of charge (SOCi) must at no hour i fall below the minimum SOCmin or exceed the maximum SoCmax at any time. Accordingly, the following condition SOCmin ≤SOCi≤SoCmax (9) applies for each hour of the optimization. Participating and trading on the DA market or the aFRR market has a direct effect on the SOCi which is represented in SOCi=SOCi−1−EnergyDASell,i+EnergyDA Buy,i−EnergyaFRR,positive,i +EnergyaFRR,negative,i(10) where the SOCi is updated every time step i. At the beginning of the optimization, the battery is half charged. Additionally, a ramping condition depicted by ∀Energyi≤SOCmax E2P(11) applies. This means, that all energy flows ∀Energyi, i.e. all purchases or sales on both markets, are subject to the maximum output rate. The energy-to-power (E2P) ratio indicates the charge or discharge in relation to its maximum capacity SOCmax. A BSS with an E2P ratio of 1 is fully charged in one hour from an empty state or can deliver full power for 1 hour, provided that it was originally fully charged. At an E2P ratio of 10, this would mean 10 h of charging or 10 h of continuous power. Therefore, the smaller the E2P ratio, the more suitable the storage is for short-term deployment. Battery degradation has not been considered in this model since the effect is expected to be very minor when interpreting the results of a single simulation year. Self-discharge has also not been considered in this work since the timescale of relevant selfdischarge is in the order of months and thus much longer than the time interval of typical battery storage use in the order of days. A binary constraint prohibits the BSS from simultaneous charging and discharging in the same hour. Finally, the optimization algorithm tries to find the best operating decision in each hour to maximize the operating result in the whole year max{RevenueTotal|conditions (9)to (11) } (12) while considering the restrictions defined in (9) to (11). We do not consider other operational costs, taxes, costs of market participation, nor prequalification costs in the presented assessment. A description of the full parameterization of all input variables to the BSS model can be found in the Appendix in Table A7. 2.3. Back-testing scenario 2019 Back-testing is an important method for evaluating the outcome of energy systems models. That is why we have set up a reference scenario for Germany in 2019 in order to compare simulated prices to historic ones. The power plant park is listed in Table 1. Despite already high shares of RE plants in Germany, electricity generation is still dominated by fossil-based generators [62]. Historic electricity prices at the DA market, load data including imports and exports as well as RE generation are derived from the SMARD data platform which is hosted by the Bundesnetzagentur [63]. European Emission Allowances were taken from the EEX [64] and used for CO 2 price information. Fuel price indices are used from the monthly reports from the Federal Statistical Office of Germany [65]. The full list of model parameters is described in the Appendix in Table A5. 2.4. Scenario 2030 In order demonstrate the feasibility of the developed approach, we decided to define a case study for Germany in 2030. However, the model set-up can also be parameterized to serve similar international electricity markets. The presented scenario follows the results of the simulations in a study on the macroeconomic effects of the energy system transformation in Germany in Lutz et al. [66]. This study aims for an Table 1 Installed power plant capacities in Germany at the end of 2018 [62] Technology GW inst Nuclear 9.5 Lignite 20.9 Hard coal 23.8 Natural gas 23.8 Other non-renewable 10.1 Pumped hydro storage 9.7 Run-of-river 3.8 Biomass 7.7 Wind onshore 50.3 Wind offshore 5.4 PV 42.3 Table 2 Installed power plant capacities in the presented scenario for Germany derived from Technology Installed Power in GW Nuclear 0 Lignite 9 Hard coal 11 Natural gas 53 Other non-renewable 5 Storage 8 Run-of-river and hydro storage 6 Biomass 6 Wind onshore 58 Wind offshore 15 PV 73 F. Nitsch et al. F. Nitsch 3 Publications 33 Applied Energy 298 (2021) 117267 6 electricity system with almost 85% CO 2 reduction in 2050 compared to 1990 and describes a pathway to this goal. The power plant park was derived from the scenario year 2030. The structure of the power plant park is listed in Table 2. There are significant capacities of photovoltaics (PV) and wind power (onshore and offshore) installed. 60% of the yearly total energy demand of 539 TWh is supplied by RE technologies. A complete nuclear-phase out is already accomplished, whereas 11 GW of hard coal and 9 GW of lignite powered plants are still in service. The price for one ton of emitted CO 2 is defined at 35 EUR/t. The market premiums for RE are assumed to be variable under the current legal framework. Accordingly, the amount of the premium is adjusted monthly according to the market values of the respective technology. An overview of all model-related assumptions and input parameters, such as fuel prices, specific emissions, power plant availability, economic factors and storage parameters can be found in the Appendix in Table A6 for the AMIRIS model and in Table A7 for the BSS optimization model. In the presented market model, no electricity transmission grid is considered. Therefore, the regulation of generation plants is only based on economic principles and not caused by the grid restrictions. The net frequencies and the required power for frequency stabilization are not simulated, but are exogenously derived from historic data. We estimated the demand for positive and negative aFRR with 1.7 GW each which is based on historic averages for the German aFRR market [67]. For the supply side, the total capacity of each technology which participates in Table 3 Capacities of technologies supplying the German aFRR market and total German aFRR demand as defined for the 2030 scenario Positive aFRR in GW Negative aFRR in GW Supply Hard coal 0.2 0.2 Lignite 0.1 0.1 Gas 3.6 3.2 Oil 0.4 0.2 Hydro power 6.6 8.4 RE (Wind/PV/ Biomass) 7.6 7.6 Demand Total power requested 1.7 1.7 Fig. 2. Comparison of simulated and historic day-ahead price-duration curves. Fig. 3. Monthly positive aFRR capacity prices, showing the median, 1st, and 3rd quartile. F. Nitsch et al. F. Nitsch 3 Publications 34 Applied Energy 298 (2021) 117267 7 the aFRR is shown in Table 3. We estimated the share of total installed capacities which can theoretically supply aFRR based on Hasche et al. [1] and for wind, PV and biomass from Spieker, Kopiske, & Tsatsaronis [68]. In the present market simulation, we have integrated an hourly bidding procedure. This is a simplification, since in reality on the German aFRR market, bids had to be submitted on a weekly, or – after changes in market design in 2018 [69] – daily basis. This adaptation had to be made in order to keep the problem solvable in the AMIRIS model. We expect, however, additional adjustments in the future which will likely introduce an even more short-term tender for the required aFRR. 3. Results The following results are divided in Section 3.1 where we describe the outcomes of the back-testing of AMIRIS whereas in Section 3.2 we present the result of the scenario for 2030. 3.1. Results back-testing scenario 2019 The simulated and historic DA prices are plotted in Fig. 2 for the year 2019 as price-duration curve. When comparing to historic prices, we observe a higher price level for simulated prices. The mean price is 38.70 EUR/MWh for the historic prices and 39.20 EUR/MWh for the Fig. 4. Annual revenues per storage capacity of the BSS operator on the DA market and aFRR market based on 2016 historic market data (left), a simulated market for 2019 (middle) and the 2030 scenario (right). Table A4 Model characteristics of AMIRIS; *D temp , D tech , D spat - temporal, technological, spatial differentiation. Model name AMIRIS - Agent-based market model for the investigation of renewable and integrated energy systems Author (Institute) German Aerospace Center (DLR), Institute of Networked Energy Systems Model type Agent based electricity market model Technical focus Electricity market, use of renewable energies under regulatory framework conditions at actor level Geographic focus Germany Spatial resolution Single bidding zone Temporal resolution Hourly Input parameters D temp D tech D spat •Costs (fixed and variable) for investment operators and direct marketers x x •Power plant efficiencies x x •General market conditions x x •Fuel prices and CO 2 certificate prices x x •Load profile x •Power plant park (conventional and renewable) x x Output parameters •Profiles of storages under different operation strategies x x •Profiles of RE plants under different regulatory frameworks x x •Electricity prices x •Revenues of direct marketers x x •Operational costs and emissions x x Table A5 Input parameters to the ABM AMIRIS in the back-testing 2019 scenario Parameter Value Unit Note Fuel prices Nuclear 3.03 EUR/ MWh Federal Statistical Office of Germany [65] Gas 27.29 EUR/ MWh Lignite 5.00 EUR/ MWh Hard coal 7.86 EUR/ MWh Oil 30.70 EUR/ MWh Specific emissions Nuclear 0 tCO 2 / MWh Gas 0.202 tCO 2 / MWh Lignite 0.364 tCO 2 / MWh Hard coal 0.341 tCO 2 / MWh Oil 0.267 tCO 2 / MWh Availabilities Nuclear 85 % Gas 97 % Lignite 98 % Hard coal 96 % Oil 93 % Minimum and maximum efficiencies Nuclear 33.0 – 33.0 % Open Power System Data [81] Gas 27.6 – 61.2 % Lignite 31.3 – 43.1 % Hard coal 28.5 – 49.0 % Oil 30.5 – 39.7 % Technical parameters of storage technologies E2P 5 h Charging efficiency 87 % Discharging efficiency 87 % Forecast period 168 h Planning period 24 h F. Nitsch et al. F. Nitsch 3 Publications 35 Applied Energy 298 (2021) 117267 8 simulated ones. Especially for lower prices, AMIRIS tends to overestimate the prices. This effect can be explained by the fact that AMIRIS does not incorporate ramping and start-up costs of power plants in a bottom-up manner. Prices in the mid-range are simulated more accurately. In hours of high demand we find a higher level of prices in the simulation compared to the historic observations. The standard deviation is 10.60 EUR/MWh in the historic case compared to 11.50 EUR/ MWh in the simulated scenario. The remaining deviations may be caused by costs for ramping power plants or due to missing depiction of block-bids in the current AMIRIS model. The correlation of the two price time series is 0.81. 3.2. Results scenario 2030 Besides the power plant dispatch, the main outputs of the AMIRIS model are the simulated price time series. Specifically, we derive a price time series for the DA market with 8760 h. The mean simulated DA price is 63 EUR/MWh. The price deviation is 12 EUR/MWh and therefore slightly higher compared to prices from 2015 to 2019 where prices deviated with 11 EUR/MWh around the mean of 35 EUR/MWh. Fig. 3 shows the capacity prices in EUR/MW for positive aFRR as boxplots for each month of the modeled scenario year 2030. We have refrained from comparing the simulated aFRR prices with 2019, as the recent changes in German aFRR market regulations mean that this is no longer viable. Instead, as described in Section 2.1, we have applied the theory according to Müsgens et al. [36]. A high variability of aFRR prices can be observed in January, February, and April; whereas especially the summer months June, July, and August have the lowest mean and additionally the smallest deviation of prices. This effect may be explained by the interplay with the DA market where lower prices are usually also found in summer. The maximum prices are around 45 EUR/ MW. Regarding the negative aFRR capacity prices and following the theory of equation (2) described in Section 2.1, we observe prices of 0 EUR/ MW. This means, that power plants with marginal costs below the forecasted DA market price can fully supply the negative aFRR capacities leading to this result. In additional calculations, we altered the required demand from currently 1.7 GW to 2 GW in a “High demand” scenario, and to 1.4 GW in a “Low demand” scenario. These variations should account for the uncertainty of future aFRR demand and their effect on prices. However, these alterations did not change the prices significantly since the capacities are sufficient to meet the demand for the aFRR. Therefore, these results are not described in more detail. While AMIRIS focuses on the whole electricity system, we can get insights in the situation for the BSS operator using our optimization model, which is described in Section 2.2. We use the price time series from the AMIRIS model as input to the optimization model and calculate the optimal BSS operation strategy. We assume that the BSS, which operates under perfect foresight, calculates its bids at the same price as the highest power plant which is still in the market. This leads to the identification of an upper limit regarding the revenue potential of the BSS operator. The results in the present scenario, however, disclose a very competitive situation on the DA market and aFRR market with overall low revenue margins for the BSS operator. Fig. 4 shows the annual revenues of a BSS on both markets with E2P ratios between 1 and 10 and a fixed roundtrip efficiency of 85% in three different situations. We compared the results from the presented scenario 2030 to revenue evaluations based on historic market data from 2016, and to the 2019 market simulations as presented in Section 3.1. The analysis for the historic market data 2016 shows the highest annual revenues. Although the power plant park has not significantly changed from 2016 to 2019, we observe reduced economic revenue potentials in the simulated market 2019. This indicates that prices on the aFRR market are probably not fully described by the theory as stated by Müsgens, Ockenfels, & Peek [36], see Section 2.1, and the simulation is Table A6 Input parameters to the ABM AMIRIS in the 2030 simulation scenario Parameter Value Unit Note Fuel prices Gas 27.0 EUR/ MWh Lutz et al. [66] Lignite 6.0 EUR/ MWh Hard coal 9.0 EUR/ MWh Oil 112.5 EUR/ MWh Specific emissions Gas 0.202 tCO 2 / MWh Lignite 0.364 tCO 2 / MWh Hard coal 0.341 tCO 2 / MWh Oil 0.267 tCO 2 / MWh Availabilities Gas 97 % Lignite 98 % Hard coal 96 % Oil 93 % Minimum and maximum efficiencies Gas 27.6 – 61.2 % Open Power System Data [81] Lignite 31.3 – 43.1 % Hard coal 28.5 – 49.0 % Oil 30.5 – 39.7 % Technical parameters of storage technologies E2P 5 h Charging efficiency 87 % Discharging efficiency 87 % Forecast period 168 h Planning period 24 h Table A7 Input parameters to the BSS optimization model Parameter Value Unit Notes Time series DA prices, historic 2016 Timeseries EUR/ MWh Bundesnetzagentur [63] DA prices, from AMIRIS Timeseries EUR/ MWh Power prices aFRR market, historic 2016 Timeseries EUR/ MW Power prices aFRR market, from AMIRIS Timeseries EUR/ MW Energy prices aFRR market, historic 2016 Timeseries EUR/ MWh Energy prices aFRR market, from AMIRIS Timeseries EUR/ MWh Technical parameters of BSS Minimum SOC 0 MWh Maximum SOC 1 MWh Initial SOC 0.5 MWh E2P 1–10 h Charging efficiency [92.20, 93.54, 94.87] % Discharging efficiency [92.20, 93.54, 94.87] % Forecast 8760 h Planning period 8760 h F. Nitsch et al. F. Nitsch 3 Publications 36 Applied Energy 298 (2021) 117267 9 probably lacking to account for strategic bidding. The distribution of revenues between the DA market and aFRR market, however, is very similar. When looking at the situation in the scenario 2030 we find that total revenues are higher compared to the simulated market 2019, but still considerably lower than in the historic market data 2016 situation. Yet, the revenues from the DA market increase strongly until 2030 because of higher fluctuations in DA prices, leading to a major shift for the primary source of revenues towards the DA market in the scenario 2030. In other words, we hardly see any significant revenues from the aFRR market in the scenario 2030 since the BSS is mainly active on the DA market. The revenue split between the DA market and aFRR market is therefore significantly different in the 2030 scenario compared to the historic case (2016 and 2019). In the latter, the revenue share on the aFRR market ranges between 77% (for E2P =1) to 27% (for E2P =10) meaning that more short-term BSS generate more revenues from providing system services such as aFRR. The picture is different for the scenario 2030 in which we observe hardly any significant revenue from the aFRR market. This is contradictory to the study by Ela et al. [70] who state that system services may become a greater proportion of revenue sources. Generally, assuming a fixed BSS capacity, the smaller the E2P ratio, the higher the expected yearly revenues. This is caused by short-term fluctuations of prices which favor short-term BSS (smaller E2P ratio). Calculations with different roundtrip efficiency levels showed that an increase in the roundtrip efficiency of one percentage point generates approximately 2.5% additional revenue for the BSS operator when operating on the DA and aFRR market. Our results show a very challenging situation for BSS operators in the future scenario for 2030. Although the BSS operator acts under perfect foresight, one cannot expect revenue opportunities as observed in 2016. This low expected profitability may lead to reduced private investments in BSS. In case BSS are identified as an essential part of future energy systems [71] investors would need access to additional, more profitable markets or require further incentives to build flexibility options, such as BSS. 4. Discussion 4.1. Limitations of the modeling approach The following points should be considered when applying the presented modeling approach and drawing conclusions from the results. First, the presented analysis does not consider all possible technologies for the provision of flexibility on the aFRR market, but uses only those listed in Table 3. For example, demand response or dispatchable loads of large consumers (e.g. industry) are not modeled. Similarly, there is neither Power-to-X nor a high penetration of electric vehicles implemented. Theoretically, these technologies expand the available capacity for frequency stabilization and could thus have an impact on prices at the DA market and aFRR market. However, they may require regulatory adaptations, which would allow them to participate at ancillary markets such as the aFRR market. Second, due to the downstream setup of the optimization model, the power supply of the modeled BSS has no influence on the coverage of the required quantity for frequency stabilization in the AMIRIS model. In the presented scenario, however, the simulated BSS has no system-relevant size. Therefore, we estimate the influence of the considered BSS operator as minimal on the change of the power prices. However, we do see the necessity of additional analyses in future work addressing the interplay of actors and their feedback on the prices. Regarding the implications of strategic bidding, Maaz [72] found that market participants add markups to their bids which can deviate from their marginal costs. Ocker, Ehrhart, & Ott [73] made an analysis of bidding strategies in the German and Austrian balancing markets, finding that the expected profits of the energy bid are taken into consideration for the calculation of the optimal power bid. Also Merten, Rücker, Schoeneberger, & Sauer [74] describe a comparison of different statistical approaches taking the acceptance probability of German aFRR bids into account. These issues may be addressed in future work to investigate the impacts on the prices and revenue potentials. Third, since the future demand for balancing power is very difficult to estimate, several variations of the required reserve power were assumed. However, only the quantity of required power, both positive and negative, was changed, but not the energy actually demanded. These quantities are difficult to project fundamentally, as they are very difficult to model and would require at least a basic implementation of the electricity grid, forecasting errors regarding load and generation as well as a representation of outages of power plants and power consumers. For these reasons, historic data on called energy is used for the analysis in this study. Fourth, the BSS in our test setup has access to the DA market and the aFRR market, as described in Section 2.2. However, BSS are also suitable for use in more short-term markets such as the Intraday market or FCR market due to their very fast response time [3]. However, these markets require a very high temporal resolution which currently cannot be modeled with AMIRIS. At the moment, we can conduct calculations on an hourly basis as described in Section 2.1. For this assessment, the aggregation of short-term markets to hourly values is not meaningful and would not achieve reasonable results. Alternatively, BSS can be active on mFRR markets competing with other large-scale power plants but also more innovative solutions such as virtual power plants or load shifting technologies in industry. The lack of these potential additional sources of income (FCR, Intraday, mFRR), however, could improve the economic situation in favor of the BSS operators. Fifth, the optimization calculation of the BSS is carried out under the assumption of perfect foresight. This means that the algorithm determining the BSS operation strategy has complete information, which is not available to this extent in reality. Therefore, the solver can calculate with market prices that will later occur exactly as expected. Additionally, we do not include taxes and levies on revenues or costs of market participation (e.g. prequalification costs for the aFRR market) for the BSS. Changes regarding the efficiency of the BSS showed no significant influence on the economic potentials. Furthermore, because cell degradation is driven primarily by calendar aging rather than cycle aging [75], we do not explicitly model this effect. In a long-term analysis of BSS, however, this has to be considered as a prominent driver in the economic evaluation. In general, we interpret the presented results as an upper-limit regarding BSS revenue potentials on the modeled markets. Finally, the lacking consideration of competition among the flexibility options should be mentioned. Such competition may lead to cannibalization effects and a further decrease of revenue potentials. The Europeanization of the electricity markets could also lead to more competition and greater pressure on individual operators in the markets and subsequently reduce the revenue opportunities of individual BSS. 4.2. Interpretation of the scenario 2030 The analysis by Braeuer et al. [10] is in line with our findings, as they conclude that investing in and operating BSS is not economically reasonable from a current point of view, despite they also considered multiple revenue possibilities. Berrada et al. [28] conducted a profit comparison between different storage technologies on DA markets and ancillary markets. Although their findings also show negative profits for innovative market participants – e.g. gravity storage – a valid comparison to our approach is not possible since they model only a single day whereas we simulate a full year. The analysis by Merten, Olk, Schoeneberger, & Sauer [76] investigates the combined use of BSS on the Intraday market and aFRR market concluding a potential economic feasibility of such systems in 2025. Angenendt, Merten, Zurmühlen, & Sauer [77] state solely the provision of frequency restoration reserve by BSS is less economical than a combined use with e.g. a PV system. The declining revenue potential for BSS over the last years is also found by Spodniak, Bertsch, & Devine [78]. Regarding different E2P ratios, the findings by He et al. [42], Engels et al. [75] and Pusceddu et al. [58] F. Nitsch et al. F. Nitsch 3 Publications 37 Applied Energy 298 (2021) 117267 10 point in similar direction by showing largest revenue potentials for short-term orientated BSS and declining profits for BSS with higher E2P ratios. As Xu et al. [11] already proposed in their study, there is a need for adapting the regulations of existing ancillary markets, such as aFRR, to compensate for the specifics of BSS. Otherwise, a market-driven integration of BSS is not likely based on current regulations. Regarding the future demand for balancing power markets in electricity systems with high shares of RE, Ocker & Ehrhart [79] addressed the questions raised by Hirth & Ziegenhagen [80], concluding that the improvement of grid control cooperation can lead to significant efficiency savings. The situation in our 2030 scenario, however, is still very unclear and we cannot model the demand endogenously. According to the prequalified capacities for the aFRR market presented in Hasche et al. [1], we assume that a rising demand for aFRR could be met without any problems, even in a scenario with reduced conventional capacities. However, since the uncertainty regarding the demand remains, we investigated the effect of different demand levels for aFRR power in a sensitivity analysis. Yet, we did not find any significant changes in prices nor in the economic potential of a BSS. This is especially true for the scenario 2030, where the prevailing share is earned at the DA market. As described in Section 3.2, the technical specifications of the BSS in form of the E2P ratio have much greater influence on its total revenues. Therefore, we renounced to alter the scenario in this regard in more detail. 5. Conclusions We present a novel approach for simulating the automatic frequency restoration reserves market alongside the day-ahead market in an agentbased electricity market model. For this purpose, we calculate bids based on the opportunity costs of market players in order to participate at the two modeled markets. First, the model was back-tested for Germany for the most recent available year 2019 achieving an overall good fit. Then, we have set up a scenario for 2030 according to a recently published study for a low-emission electricity system in Germany. The simulated electricity system features a significant share of renewable power plants supplying already 60% of the yearly electricity demand. From this scenario and model setup, we derive price time series for both investigated markets. We then assess the revenue potentials of battery storage system operators which are active on these two markets. In an optimization model, we calculate the optimal storage dispatch strategy and evaluate its profitability. When we compare the simulated potential revenues in the given scenario 2030 to those revenues in a simulated market 2019, we see an improved economic potential in the simulated future scenario. Additionally, in the scenario 2030 the distribution of revenues shifts towards the day-ahead market which is explained by higher price fluctuations. The technical specifications of the battery storage system are crucial for an optimal use-case. We find that the ability to provide power in the short-term leads to the highest revenues concluding that high power battery storage systems perform best in the given scenarios. Higher round-trip efficiency only contributes to minor improvements regarding the annual revenues. Additional calculations could further enhance the presented results by taking the investment and operational costs of battery storage systems into account. Future work may also improve the presented approach by including additional markets such as the Frequency Containment Reserves market or Intraday market into the model to generate a more comprehensive view of the revenue potentials. As discussed, the battery storage system operator may increase its revenue when employing a multi-use strategy to serve various markets simultaneously. While the presented modeling approach is demonstrated for the specifications of the German market, the developed methodology can be adapted to describe the situation on different national electricity markets such as North America or European countries. This enables policy makers, companies, and investors to get a better understanding of the application of battery storage systems. For this purpose, technical specifications may have to be adjusted to reflect the corresponding market design and rules. In addition, the regionspecific power plant parks and market prequalification requirements to participate in the day-ahead market and frequency restoration reserves market must be considered accordingly. Credit authorship contribution statement Felix Nitsch: Conceptualization, Methodology, Software, Validation, Investigation, Visualization, Writing - original draft. Marc Deissenroth-Uhrig: Conceptualization, Methodology, Software, Writing - review & editing. Christoph Schimeczek: Validation, Investigation, Writing - review & editing. Valentin Bertsch: Writing - review & editing, Supervision. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements The authors thank Kristina Nienhaus, Benjamin Fuchs, and the colleagues from the Department of Energy Systems Analysis at the Institute of Networked Energy Systems, German Aerospace Center (DLR), for their valuable comments and fruitful discussions on early versions of the manuscript. We gratefully acknowledge support from the German Federal Ministry for Economic Affairs and Energy (grant number: 03ET4032). Appendix References [1] Hasche, B., Haiges, M., & Schulz, S. (2016). Die Integration von Batteriespeichern am Regelleistungsmarkt - eine modellgestützte Bewertung. 50Hertz. [2] Ehrhart, K.-M., Belica, M., & Ocker, F. (2016). Positionspapier: Umstrukturierung des deutschen Sekund¨ arregelmarktes für Strom. [3] Denholm, P., Ela, E., Kirby, B., & Milligan, M. (2010). The role of energy storage with renewable electricity generation. National Renewable Energy Laboratory. Retrieved from http://www.nrel.gov/docs/fy10osti/47187.pdf. [4] Hollinger R, Diazgranados LM, Erge T. In: Trends in the German PCR market: Perspectives for battery systems. IEEE; 2015. p. 1–5. [5] Commission European. COMMISSION REGULATION (EU) 2017/2195 of 23 November 2017 establishing a guideline on electricity balancing. Official Journal of the European Union, L 2017;312(6):1–48. [6] Ocker F, Ehrhart K-M, Belica M. Harmonization of the European balancing power auction: A game-theoretical and empirical investigation. Energy Econ 2018;73: 194–211. [7] Thorbergsson, E., Knap, V., Swierczynski, M., Stroe, D., & Teodorescu, R. (2013). Primary frequency regulation with Li-ion battery based energy storage systemevaluation and comparison of different control strategies. Intelec 2013; 35th International Telecommunications Energy Conference, SMART POWER AND EFFICIENCY (pp. 1–6). VDE. [8] Fleer J, Stenzel P. Impact analysis of different operation strategies for battery energy storage systems providing primary control reserve. J Storage Mater 2016;8: 320–38. [9] Zeh, A., Mueller, M., Hesse, H. C., Jossen, A., & Witzmann, R. (2015). Operating a multitasking stationary battery storage system for providing secondary control reserve on low-voltage level. International ETG Congress 2015; Die EnergiewendeBlueprints for the new energy age (pp. 1–8). VDE. [10] Braeuer F, Rominger J, McKenna R, Fichtner W. Battery storage systems: An economic model-based analysis of parallel revenue streams and general implications for industry. Appl Energy 2019;239:1424–40. [11] Xu B, Oudalov A, Poland J, Ulbig A, Andersson G. BESS control strategies for participating in grid frequency regulation. IFAC Proceedings Volumes 2014;47(3): 4024–9. [12] Tian Y, Bera A, Benidris M, Mitra J. Stacked revenue and technical benefits of a grid-connected energy storage system. IEEE Trans Ind Appl 2018;54(4):3034–43. [13] Vejdan S, Grijalva S. In: Analysis of multiple revenue streams for privately-owned energy storage systems. IEEE; 2018. p. 1–5. F. Nitsch et al. F. Nitsch 3 Publications 38 Applied Energy 298 (2021) 117267 11 [14] Hu J, Sarker MR, Wang J, Wen F, Liu W. Provision of flexible ramping product by battery energy storage in day-ahead energy and reserve markets. IET Gener Transm Distrib 2018;12(10):2256–64. [15] Borsche T, Ulbig A, Koller M, Andersson G. In: Power and energy capacity requirements of storages providing frequency control reserves. IEEE; 2013. p. 1–5. [16] Doherty R, Lalor G, O’Malley M. Frequency control in competitive electricity market dispatch. IEEE Trans Power Syst 2005;20(3):1588–96. [17] Ela, E., Kirby, B., Botterud, A., Milostan, C., Krad, I., & Koritarov, V. (2013). Role of pumped storage hydro resources in electricity markets and system operation. National Renewable Energy Lab.(NREL), Golden, CO (United States). [18] Kirby, B. J., & Kueck, J. D. (2005). Spinning reserve from pump load: A technical findings report to the California Department of Water Resources. Department of Energy. [19] O’Sullivan, J., Power, M., Flynn, M., & O’Malley, M. (1999). Modelling of frequency control in an island system. IEEE Power Engineering Society. 1999 Winter Meeting (Cat. No. 99CH36233) (Vol. 1, pp. 574–579). IEEE. [20] Wu C-C, Lee W-J, Cheng C-L, Lan H-W. In: Role and value of pumped storage units in an ancillary services market for isolated power systems-simulation in the Taiwan power system. IEEE; 2007. p. 1–6. [21] Parag Y, Sovacool BK. Electricity market design for the prosumer era. Nat Energy 2016;1(4):1–6. [22] Iria J, Soares FJ, Matos MA. Trading small prosumers flexibility in the energy and tertiary reserve markets. IEEE Trans Smart Grid 2018;10(3):2371–82. [23] Iria J, Soares F, Matos M. Optimal bidding strategy for an aggregator of prosumers in energy and secondary reserve markets. Appl Energy 2019;238:1361–72. [24] Bessa RJ, Matos MA. Optimization models for EV aggregator participation in a manual reserve market. IEEE Trans Power Syst 2013;28(3):3085–95. [25] Bessa RJ, Matos M. Optimization models for an EV aggregator selling secondary reserve in the electricity market. Electr Power Syst Res 2014;106:36–50. [26] Vatandoust B, Ahmadian A, Golkar MA, Elkamel A, Almansoori A, Ghaljehei M. Risk-averse optimal bidding of electric vehicles and energy storage aggregator in day-ahead frequency regulation market. IEEE Trans Power Syst 2018;34(3): 2036–47. [27] Gonz´ alez-Garrido A, Saez-de-Ibarra A, Gazta˜ naga H, Milo A, Eguia P. Annual Optimized Bidding and Operation Strategy in Energy and Secondary Reserve Markets for Solar Plants with Storage Systems. IEEE Trans Power Syst 2018;34(6): 5115–24. [28] Berrada A, Loudiyi K, Zorkani I. Valuation of energy storage in energy and regulation markets. Energy 2016;115:1109–18. [29] Chazarra, M., Garcɩa-Gonz´ alez, J., P´ erez-Dɩaz, J. I., & Arteseros, M. (2016). Stochastic optimization model for the weekly scheduling of a hydropower system in day-ahead and secondary regulation reserve markets. Electric Power Systems Research, 130, 67–77. [30] Khorasani J, Mashhadi HR. Bidding analysis in joint energy and spinning reserve markets based on pay-as-bid pricing. IET Gener Transm Distrib 2012;6(1):79–87. [31] Swider DJ, Weber C. Bidding under price uncertainty in multi-unit pay-as-bid procurement auctions for power systems reserve. Eur J Oper Res 2007;181(3): 1297–308. [32] Swider, D. J. (2005). Sequential bidding in day-ahead auctions for spot energy and power systems reserve. Proceedings of the 7th IAEE European Conference on“ Energy Markets in Transition”. Bergen. [33] Swider DJ. Simultaneous bidding in day-ahead auctions for spot energy and power systems reserve. Int J Electr Power Energy Syst 2007;29(6):470–9. [34] Chao H-P, Wilson R. Multi-dimensional procurement auctions for power reserves: Robust incentive-compatible scoring and settlement rules. J Regul Econ 2002;22 (2):161–83. [35] Mazzi N, Kazempour J, Pinson P. Price-taker offering strategy in electricity pay-asbid markets. IEEE Trans Power Syst 2017;33(2):2175–83. [36] Müsgens F, Ockenfels A, Peek M. Economics and design of balancing power markets in Germany. Int J Electr Power Energy Syst 2014;55:392–401. [37] Loesch M, Rominger J, Nainappagari S, Schmeck H. In: Optimizing Bidding Strategies for the German Secondary Control Reserve Market: The Impact of Energy Prices. IEEE; 2018. p. 1–5. [38] Fleer J, Zurmühlen S, Meyer J, Badeda J, Stenzel P, Hake J-F, et al. Price development and bidding strategies for battery energy storage systems on the primary control reserve market. Energy Procedia 2017;135:143–57. [39] Wen F, David A. Coordination of bidding strategies in day-ahead energy and spinning reserve markets. Int J Electr Power Energy Syst 2002;24(4):251–61. [40] Ottesen SØ, Tomasgard A, Fleten S-E. Multi market bidding strategies for demand side flexibility aggregators in electricity markets. Energy 2018;149:120–34. [41] Goebel C, Hesse H, Schimpe M, Jossen A, Jacobsen H-A. Model-based dispatch strategies for lithium-ion battery energy storage applied to pay-as-bid markets for secondary reserve. IEEE Trans Power Syst 2016;32(4):2724–34. [42] He G, Chen Q, Kang C, Pinson P, Xia Q. Optimal bidding strategy of battery storage in power markets considering performance-based regulation and battery cycle life. IEEE Trans Smart Grid 2015;7(5):2359–67. [43] Jia Y, Mi Z, Yu Y, Song Z, Sun C. A bilevel model for optimal bidding and offering of flexible load aggregator in day-ahead energy and reserve markets. IEEE Access 2018;6:67799–808. [44] Aldik A, Al-Awami AT, Alismail F. In: Fuzzy optimization-based sizing of a battery energy storage system for participating in ancillary services markets. IEEE; 2018. p. 1–7. [45] Attaviriyanupap P, Kita H, Tanaka E, Hasegawa J. New bidding strategy formulation for day-ahead energy and reserve markets based on evolutionary programming. Int J Electr Power Energy Syst 2005;27(3):157–67. [46] Cheng B, Powell WB. Co-optimizing battery storage for the frequency regulation and energy arbitrage using multi-scale dynamic programming. IEEE Trans Smart Grid 2016;9(3):1997–2005. [47] Tesfatsion L. Agent-based computational economics: A constructive approach to economic theory. In: Tesfatsion L, Judd KL, editors. Handbook of computational economics, volume 2. Amsterdam: Elsevier North-Holland; 2006. [48] Deissenroth M, Klein M, Nienhaus K, Reeg M. Assessing the Plurality of Actors and Policy Interactions: Agent-Based Modelling of Renewable Energy Market Integration. Complexity 2017;2017. https://doi.org/10.1155/2017/7494313. [49] van Stiphout A, De Vos K, Deconinck G. In: Operational flexibility provided by storage in generation expansion planning with high shares of renewables. IEEE; 2015. p. 1–5. [50] Alqurashi A, Etemadi AH, Khodaei A. Treatment of uncertainty for next generation power systems: State-of-the-art in stochastic optimization. Electr Power Syst Res 2016;141:233–45. [51] Wierzbowski M, Lyzwa W, Musial I. MILP model for long-term energy mix planning with consideration of power system reserves. Appl Energy 2016;169:93–111. [52] Belderbos, A., Virag, A., D’haeseleer, W., & Delarue, E. (2017). Considerations on the need for electricity storage requirements: Power versus energy. Energy Conversion and Management, 143, 137–149. [53] Limpens G, Moret S, Jeanmart H, Mar´ echal F. EnergyScope TD: A novel opensource model for regional energy systems. Appl Energy 2019;255:113729. [54] Huang Q, Xu Y, Courcoubetis C. Stackelberg competition between merchant and regulated storage investment in wholesale electricity markets. Appl Energy 2020; 264:114669. [55] Shabani M, Dahlquist E, Wallin F, Yan J. Techno-economic comparison of optimal design of renewable-battery storage and renewable micro pumped hydro storage power supply systems: A case study in Sweden. Appl Energy 2020;279:115830. [56] Rodrigues DL, Ye X, Xia X, Zhu B. Battery energy storage sizing optimisation for different ownership structures in a peer-to-peer energy sharing community. Appl Energy 2020;262:114498. [57] Huang Q, Xu Y, Courcoubetis C. Financial incentives for joint storage planning and operation in energy and regulation markets. IEEE Trans Power Syst 2019;34(5): 3326–39. [58] Pusceddu E, Zakeri B, Gissey GC. Synergies between energy arbitrage and fast frequency response for battery energy storage systems. Appl Energy 2021;283: 116274. [59] Reeg M, Nienhaus K, Roloff N, Pfenning U, Deissenroth M, Wassermann S, et al. AMIRIS - Weiterentwicklung eines agentenbasierten Simulationsmodells zur Untersuchung des Akteursverhaltens bei der Marktintegration von Strom aus erneuerbaren Energien unter verschiedenen F¨ ordermechanismen. DLR - Deutsches Zentrum für Luft und Raumfahrt, IZES. Institut für ZukunftsEnergiesysteme, Kast Simulation Solution; 2013. [60] Rosenthal RE. GAMSa user’s guide. ALS-NSCORT. 2012. [61] IBM. (2017). IBM ILOG CPLEX Optimization StudioCPLEX User’s Manual. IBM Corporation. Retrieved from https://www.ibm.com/support/knowledgecenter/ SSSA5P_12.8.0/ilog.odms.studio.help/pdf/usrcplex.pdf?origURL=SSSA5P_12.8.0/ ilog.odms.studio.help/Optimization_Studio/topics/PLUGINS_ROOT/ilog.odms. studio.help/pdf/usrcplex.pdf. [62] Bundesnetzagentur. (2019). Kraftwerksliste der Bundesnetzagentur. Retrieved from https://www.bundesnetzagentur.de/DE/Sachgebiete/ElektrizitaetundGas/ Unternehmen_Institutionen/Versorgungssicherheit/Erzeugungskapazitaeten/ Kraftwerksliste/kraftwerksliste-node.html. [63] Bundesnetzagentur. (2020). SMARD - German electricity market data platform. Retrieved from https://www.smard.de. [64] EEX. (2020). Emission Spot Primary Market Auction Report 2019. European Energy Exchange AG. Retrieved from https://www.eex.com/dl/en/market-data/ environmental-markets/auction-market/european-emission-allowances-auction/ european-emission-allowances-auction-download/92076/file. [65] Federal Statistical Office of Germany. (2020). Data on energy price trends - Longtime series from January 2005 to March 2020. Federal Statistical Office of Germany. Retrieved from https://www.destatis.de/EN/Themes/Economy/Prices/ Publications/Downloads-Energy-Price-Trends/energy-price-trends-pdf-5619002. pdf?__blob=publicationFile. [66] Lutz, C., Flaute, M., Lehr, U., Kemmler, A., Kirchner, A., Maur, A. auf der, Ziegenhagen, I., et al. (2018). Gesamtwirtschaftliche Effekte der Energiewende. GWS, Fraunhofer ISI, DIW, DLR, Prognos. [67] 50Hertz, Amprion, TenneT, & TransnetBW. (2020a). Regelleistung.net - Datencenter. Regelleistung.net - Internetplattform zur Vergabe von Regelleistung. [68] Spieker, S., Kopiske, J., & Tsatsaronis, G. (2016). Flexibilit¨ at aus Windund Photovoltaikanlagen im Regelenergiemarkt 2035. 14. Symposium Energieinnovation. TU Graz. Retrieved from https://www.energietechnik.tuberlin.de/fileadmin/fg106/Dateien/Mitarbeiter/Spieker_Kopiske_EnInnov2016_ Flexibilitaet_aus_Wind-_und_Photovoltaikanlagen_im_Regelenergiemarkt_2035__ Langfassung_.pdf. [69] 50Hertz, Amprion, TenneT, & TransnetBW. (2020b). FAQ Regelleistung. Regelleistung.net - Internetplattform zur Vergabe von Regelleistung. Retrieved from https://www.regelleistung.net/ext/download/key_SRL-Abr-FAQ. [70] Ela E, Billimoria F, Ragsdale K, Moorty S, O’Sullivan J, Gramlich R, et al. Future Electricity Markets: Designing for Massive Amounts of Zero-Variable-Cost Renewable Resources. IEEE Power Energ Mag 2019;17(6):58–66. [71] Weitemeyer S, Kleinhans D, Vogt T, Agert C. Integration of Renewable Energy Sources in future power systems: The role of storage. Renewable Energy 2015;75: 14–20. [72] Maaz, A. (2017). Auswirkungen von strategischem Bietverhalten auf die Marktpreise am deutschen Day-Ahead-Spotmarkt und an den F. Nitsch et al. F. Nitsch 3 Publications 39 Applied Energy 298 (2021) 117267 12 Regelleistungsauktionen. RWTH Aachen University. Retrieved from https:// publications.rwth-aachen.de/record/692322. [73] Ocker F, Ehrhart K-M, Ott M. Bidding strategies in Austrian and German balancing power auctions. Wiley Interdisciplinary Reviews: Energy and Environment 2018;7 (6):1–16. [74] Merten M, Rücker F, Schoeneberger I, Sauer DU. Automatic frequency restoration reserve market prediction: Methodology and comparison of various approaches. Appl Energy 2020;268:114978. [75] Engels J, Claessens B, Deconinck G. Techno-economic analysis and optimal control of battery storage for frequency control services, applied to the German market. Appl Energy 2019;242:1036–49. [76] Merten M, Olk C, Schoeneberger I, Sauer DU. Bidding strategy for battery storage systems in the secondary control reserve market. Appl Energy 2020;268:114951. [77] Angenendt G, Merten M, Zurmühlen S, Sauer DU. Evaluation of the effects of frequency restoration reserves market participation with photovoltaic battery energy storage systems and power-to-heat coupling. Appl Energy 2020;260: 114186. [78] Spodniak P, Bertsch V, Devine M. The Profitability of Energy Storage in European Electricity Markets. The Energy Journal 2021;42(5). https://doi.org/10.5547/ 01956574.42.5.pspo. [79] Ocker F, Ehrhart K-M. The “German Paradox” in the balancing power markets. Renew Sustain Energy Rev 2017;67:892–8. [80] Hirth L, Ziegenhagen I. Balancing power and variable renewables: Three links. Renew Sustain Energy Rev 2015;50:1035–51. [81] Open Power System Data. (2018). Data Package Conventional power plants. Version 2018-12-20. https://data.open-power-system-data.org/conventional_ power_plants/2018-12-20 (Primary data from various sources, for a complete list see URL). Open Power System Data. Retrieved from https://doi.org/10.25832/ conventional_power_plants/2018-12-20. F. Nitsch et al. F. Nitsch 3 Publications 40 F. Nitsch 3.2 Paper II: The future role of Carnot batteries in Central Europe: Combining energy system and market perspective Authors: Felix Nitsch, Manuel Wetzel, Hans Christian Gils, Kristina Nienhaus Corresponding Author: Felix Nitsch Journal: Journal of Energy Storage Volume: 85 DOI: 10.1016/j.est.2024.110959 Status: Published 28 February 2024 Licence: Open Access, CC BY-NC-ND 4.0 Executive Summary: Carnot batteries, high-temperature heat storage systems, promise an attractive solution to meet growing flexibility needs systems with high RE shares. To assess the future economic potential of such Carnot batteries, I coupled the energy systems optimisation model REMix with the agent-based electricity market model AMIRIS.REMix evaluates the least-cost infrastructure configuration of the energy system, while AMIRIS focuses on the profitability of these storage systems. I applied this modelling chain in a case study of a zero-emission energy system in Central Europe for the year 2050. To guide the development of this promising technology, I conducted a parameter scan of costs and efficiencies of Carnot batteries in this system. My findings show that the availability of a low-cost storage medium is a key driver for the use of Carnot batteries from an energy system design perspective. Additionally, combining Carnot batteries with wind energy offers benefits as a result of the potential for longer storage durations compared to electrochemical batteries. Carnot battery operators could potentially realise positive annual gross profits, depending on the system design, their role within the energy system, and importantly, their market power and bidding strategy. On the basis of these results, I conclude that to make Carnot batteries competitive with other storage technologies on a broader scale, their development potential must be fully leveraged. This work builds on the preliminary study that was published as part of the proceedings to the International Conference on Applied Energy 2022 in Bochum, Germany (Nitsch and Wetzel 2022). Author Contributions: I am the lead author of this paper. The main research work of this paper was shared between Manuel Wetzel, a PhD student and researcher at DLR, and myself. Manuel Wetzel carried out the work on the energy system optimisation model REMix, while I carried out the work on the agent-based electricity market model 3 Publications 41 F. Nitsch AMIRIS. I also ensured the coupling of the two models through a dedicated workflow. The analysis and visualisation were done by me and Manuel Wetzel. The original manuscript was written by me and Manuel Wetzel. The manuscript was reviewed and edited by Hans Christian Gils and Kristina Nienhaus, both group leaders and senior scientists at DLR. Hans Christian Gils and Kristina Nienhaus were also responsible for the supervision and acquisition of funding. 3 Publications 42 Journal of Energy Storage 85 (2024) 110959 7 prevalent in the Low Flex and No Grid scenarios. Notably, all strategies profit from employing a rolling window of perfect foresight for arbitrage options. Even though the Carnot battery capacity in the No Grid scenario is more than doupled compared to the Low Flex scenario, it cannot outperform the gross profits from the latter. We observe diminishing spreads in electricity prices as a consequence of arbitrage. Therefore, the increased trading capacities of the Carnot battery in the No Grid scenario cannot generate additional revenue potentials. Table 2 provides additional comparative evaluation of the Maximize Profits and Minimize System Costs strategies. In this analysis, values exceeding 100 % indicate a greater impact when employing the Maximize Profits strategy. The Maximize Profits strategy also significantly influences mean prices, driving them up by at least 345 % in the Base scenario and as much as almost 400 % in the Low Flex scenario. Full cycles tend to be lower compared when aiming at maximizing profits. Total system costs (sum of all operational costs) are more than doubled (Base and Low Flex) or even tripled (No Grid) scenario. Regarding accumulated discharged and charged energy, the results reveal higher values in the Base scenario, contrasting with smaller values in the Low Flex and No Grid scenarios. 4. Discussion The results show that with our model setup and scenarios analyzed, Carnot batteries have a limited role in the modelled optimal future energy systems for Central Europe, even with optimistic cost assumptions. This results from the extensive provision of flexibility through sector coupling technologies, such as flexible hydrogen production or advanced district heating, and from the use of battery storage, which proves to be more cost-effective for many locations. However, further development, especially based on Brayton cycles and Rankine cycles in combination with heat pumps, can make Carnot batteries a promising alternative for electricity storage. Though, the future role of Carnot batteries will likewise depend on the future development of battery storage systems and electrolysers. Both technologies can have a significant impact on the overall landscape of flexibility options. This balance may be shifted if additional factors, such as material availability or increasing prices for raw materials are considered. Therefore, additional research on the life cycle impacts of different storage technologies will be an important field of research going forward. The REMix parameterization used here considers the power grid only in aggregate form as transmission capacities between model regions (Fig. 8). As a result, information about grid congestion within these regions is lost. Consequently, flexibility needs at the local level are partially underestimated, and so are the potentials of Carnot batteries at locations of high generation surpluses. The extent to which local wind power curtailments can be cost-effectively avoided by Carnot batteries Fig. 5. Hourly feed-in from renewable technologies (top), hourly charging and discharging (middle) for batteries and Carnot batteries (middle), and storage levels (bottom) throughout the year for Carnot batteries and lithium-ion batteries. Hourly values are derived from the techno-economical configuration of 65 % round-trip efficiency, 20 € /kWh storage specific CAPEX and 270 € /kW power specific CAPEX. Fig. 6. Installed Carnot battery capacities and their Energy to Power ratio (red framed crosses). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) F. Nitsch et al. F. Nitsch 3 Publications 49 Journal of Energy Storage 85 (2024) 110959 8 thus remains to be addressed in more spatially detailed analyses. The electricity market analysis of our study focuses on the economic analysis of Carnot batteries on the German market in future scenarios. [54] previously explored the economic viability of pumped heat electric storage on historical 2016 day-ahead prices concluding that high investment costs posed challenges for profitability. [55] simulated a Carnot battery on the scale of multiple households achieving similar results. In contrast, our results indicate positive gross profits, although with optimistic learning rates regarding CAPEX. Additionally, our work expands its scope beyond residential applications and considers Carnot batteries at a larger scale, with variations in power and capacity configurations. [56] explored the optimal sizing of Carnot batteries in combination with concentrated solar power plants, focusing on historical day-ahead prices in the Spanish electricity market and identifying E2P ratios between 5 and 10 as optimal for intraday storage purposes. Our study also suggests E2P ratios between 7.4 and 8.2 resulting in similar characteristics. Furthermore, in line with [57], who emphasized the significance of E2P ratios greater than 7 and steep electricity price increases for Carnot battery applications, our findings support the importance of the scaling of the Carnot battery for achieving profitability. In [19], the authors investigate a 100 % renewable energy system proposed for a future Danish energy system in 2045. The findings reveal that Carnot batteries have the potential to facilitate 32 annual storage cycles, in combination with significantly elevated E2P ratios. Consistent with our own results, the authors underscore the importance of decreasing storage costs to thresholds below 60.5 € /MWh el and 38 € /MWhel, contingent upon the specific sub-scenario considered. When interpreting the presented results on economic perspectives the following limitations have to be considered. First, the technical assumptions and cost basis of the presented Carnot batteries follow very optimistic learning rates. The assumed storage power and energy cost assumptions of 150 EUR/kW and 20 EUR/kWh, respectively, combined with a round-trip efficiency of 65 % must be kept in mind when interpreting the market analysis results. The benefit from a more integrated system such as thermal integration of Carnot batteries with industry processes or district heating systems as well as retrofitting power plants to storage systems, or competition with other flexibility options are outside the scope of this study and may shift the conclusions on the overall energy system design and the economic profitability. Second, the scope of the electricity market analysis is limited to arbitrage trading on the German day-ahead market. This not only neglects additional revenue potentials like providing system services such as frequency restoration reserve, but also possible competition from neighbouring market zones and other flexibility providers. Third, we want to emphasize that the profitability of the Maximize profits strategy marks the most upper limit of possible revenues since the storage trader benefits from its total market power and makes full use of it. 5. Conclusions We present a comprehensive analysis of Carnot batteries and assess their future role in energy systems with high shares of renewable energies. A model coupling of the energy system optimization model REMix with the electricity market simulation model AMIRIS allows an investigation on both system and market perspective. From a general energy system design perspective, we can conclude that Carnot batteries Fig. 7. Relative gross profit per MW installed compared to best performing combination (Low Flex with Maximize Profits strategy). Table 2 Evaluation of the performance of the Maximize Profits strategy in comparison to the Minimize System Costs strategy, where a value greater than 100 % indicates a more pronounced impact when applying the Maximize Profits strategy. Total System Cost Mean Price Full Cycles Accumulated Discharged Energy Accumulated Charged Energy Base 235 % 345 % 85 % 85 % 84 % No Grid 309 % 378 % 81 % 81 % 81 % Low Flex 262 % 395 % 80 % 80 % 80 % F. Nitsch et al. F. Nitsch 3 Publications 50 Journal of Energy Storage 85 (2024) 110959 9 may be a promising option for mid-term energy storage if technology development makes significant progress. In terms of system parameters this translates into the need for achieving a low-cost storage medium in the range of 20–35 € /kWh. Improving the round-trip efficiency seems to be a viable secondary target, however, needs to be traded-off against increases in the capital expenditures for charging and discharging, which may increase accordingly. If this is achieved, Carnot batteries would be more viable for high energy to power ratios than lithium-ion battery storage systems. The results from the REMix model further indicate synergies between Carnot batteries in conjunction with electricity generation from wind turbines but also to a certain degree with photovoltaic and lithium-ion battery systems. However, this synergy depends on the overall need for energy storage which can be impacted by high shares of electrolysis. The results also confirm that the use of flexible sector coupling, realized through storage for heat and hydrogen, reduces the demand for electricity storage. This has a noticeable impact on the market potential for Carnot batteries. Regarding the profitability analysis, we simulate the German day-ahead market using AMIRIS identifying positive gross profits among different scenarios. We conclude that the gross profit of Carnot battery storage systems is highly impacted by their considerable size and their favourable market position. Our results indicate that profitability is strongly related to market power of the storage operator which is particularly pronounced when the profit maximization strategy is applied. Therefore, further research may focus on a more accurate simulation of the competition of flexibility options and on finding robust strategies for the storage operator considering its impact of market power. Additional revenue streams such as ancillary markets could also be integrated in upcoming studies. Additionally, we propose that future investigations should extend the scope to other regions worldwide, recognizing the potential variability in the energy landscape and market dynamics, thereby contributing to a more comprehensive understanding of Carnot battery applications on a global scale. Author agreement statement We declare that this manuscript is original, has not been published before and is not currently being considered for publication elsewhere. We confirm that the manuscript has been read and approved by all named authors and that there are no other persons who satisfied the criteria for authorship but are not listed. We further confirm that the order of authors listed in the manuscript has been approved by all of us. We understand that the Corresponding Author, Felix Nitsch, is the sole contact for the Editorial process. He is responsible for communicating with the other authors about progress, submissions of revisions and final approval of proofs. CRediT authorship contribution statement Felix Nitsch: Writing – original draft, Visualization, Software, Methodology, Investigation, Conceptualization. Manuel Wetzel: Writing – original draft, Visualization, Software, Methodology, Investigation, Conceptualization. Hans Christian Gils: Writing – review & editing, Supervision, Funding acquisition, Conceptualization. Kristina Nienhaus: Writing – review & editing, Supervision, Funding acquisition, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Data availability Data will be made available on request. Acknowledgements The research for this paper was conducted within the “CarnotBat” project, which was designed and realized in co-operation with other institutes of the German Aerospace Center (DLR). The work presented here was funded by the “Energy Systems Design” programme of the Helmholtz Association. The authors thank their colleagues from the Department of Energy Systems Analysis at the Institute of Networked Energy Systems, German Aerospace Center (DLR), for their valuable comments and fruitful discussions on early versions of the manuscript. Appendix A. Overview of the model workflows Fig. 8. Schematic overview of the REMix energy system model, from [58]. F. Nitsch et al. F. Nitsch 3 Publications 51 Journal of Energy Storage 85 (2024) 110959 10 Fig. 9. Schematic overview of the electricity market model AMIRIS, see also [28]. Appendix B. Sensitivity analysis for the energy system design As described in Section 2.3, further model calculations were carried out with REMix to examine how deviating assumptions on the composition of electricity generation and storage technology development affect the role of Carnot batteries in the cost-optimal system. This is realized by varying the cost assumptions for lithium-ion battery storage, P2G2P, photovoltaics and wind power plants. The additional assumptions used are based on values collected by the Danish Energy Agency [59] and the ranges of investment costs in 2050 mentioned therein. The resulting assumptions for battery storage costs are summarized in Table 3 and the assumptions for wind energy, photovoltaics and P2G2P in Table 4. The sensitivity analyses are carried out for the following three cases from the parameter study of technoeconomic assumptions on Carnot battery systems. (1) 55 % round-trip efficiency, 35 € /kWh energy-specific costs and 150 € /kW power-specific costs (2) 75 % round-trip efficiency, 35 € /kWh energy-specific costs and 400 € /kW power-specific costs (3) 75 % round-trip efficiency, 75 € /kWh energy-specific costs and 150 € /kW power-specific cost Table 3 Lithium-ion battery cost assumptions in the sensitivity analysis. The Base case values represent the assumptions used in the parametric study with the results presented in Section 3.1. base battery++ battery+battery0 batterybattery– Energy storage expansion cost ( € /kWh) 75 46 78.5 111 143.5 176 Output capacity expansion cost ( € /kW) 60 40 92.5 145 197.5 250 Table 4 Cost assumptions for wind energy, photovoltaics, electrolysis and methanation in the sensitivity analysis. If a field specifies no values the Base values are used. For all technologies fixed operational costs are scaled accordingly. base p2g2p+p2g2ppv +windpv +wind+pv-wind+ Electrolyzer expansion cost ( € /kW) 350 150 500 Methanizer expansion cost ( € /kW) 800 450 Photovoltaic expansion cost ( € /kW) 518 250 250 Onshore wind expansion cost ( € /kW) 1173 800 800 Offshore wind expansion cost ( € /kW) 1800 1640 1640 F. Nitsch et al. F. Nitsch 3 Publications 52 Journal of Energy Storage 85 (2024) 110959 11 Fig. 10. Discharged energy across Europe (battery scenarios). Varying the costs for lithium-ion batteries results in the expected effects, see Fig. 10. Thus, with higher battery costs in all three cases examined, there is almost complete substitution of lithium-ion batteries by Carnot batteries. The total capacity of the battery storage systems remains approximately constant. The effects are uniform for the three sets of assumptions analyzed for Carnot batteries. Assuming lower costs for lithium-ion storage systems, on the other hand, Carnot batteries are completely pushed out of the system and the total capacity of the storage systems is doubled or tripled. As no Carnot batteries are used anymore, the difference between the three model runs disappears. Fig. 11. Annual electricity generation across Europe (battery scenarios). The variation in battery costs has only a minor impact on the power generation structure (Fig. 11). These are most evident in the case of lower battery costs, which lead to CSP and partly also onshore wind being replaced by PV. Higher battery costs, on the other hand, lead to a slight increase in total electricity production, as the use of Carnot batteries is associated with higher losses. F. Nitsch et al. F. Nitsch 3 Publications 53 Journal of Energy Storage 85 (2024) 110959 12 Fig. 12. Discharged energy across Germany (battery scenarios, without methanizer). The described effect of the cost variations on the entire study area is essentially also confirmed for Germany. However, the importance of electricity storage is lower there due to the greater availability of other flexibility. This means that an increase in lithium-ion battery costs only has a very insignificant effect on the use of electricity storage (Fig. 12), although in case of a significant cost increase (battery–), lithium-ion batteries are replaced by Carnot batteries. A reduction in the cost of lithium-ion batteries, on the other hand, would mean that they would find a place in the German system and significantly increase the importance of electricity storage. Fig. 13. Annual electricity generation in Germany (battery scenarios). The analysis of electricity generation in Germany shows that the variation in lithium-ion battery storage costs only changes this very slightly (Fig. 13). The most relevant aspect is the slight decrease in total generation due to the reduction in renewable energy curtailment and storage losses in case of significantly cheaper chemical batteries (battery++). F. Nitsch et al. F. Nitsch 3 Publications 54 Journal of Energy Storage 85 (2024) 110959 13 Fig. 14. Discharged energy across Europe (VRE scenarios). The considered variations in the cost assumptions for the production and reconversion of synthetic methane only have a very weak effect on the results (Fig. 14). As the methanization plants are essentially used to cover gas demand in industry, cost changes have hardly any influence. This also applies to the other storage technologies analyzed. A different picture emerges when varying the costs of VRE technologies. Reduced PV costs significantly increase the contribution of this technology to electricity generation and push offshore wind in particular out of the system (Fig. 15). This results in a significantly higher storage requirement, which is covered disproportionately by Carnot batteries. Discharge from lithium-ion batteries also doubles. If both wind and PV costs are assumed to be lower, this has a particular impact on wind power generation, where offshore wind is replaced by onshore wind. This also results in a change in storage requirements. Although this hardly increases for the sum of lithium-ion and Carnot batteries, the latter can significantly increase their share. If a cost reduction is only assumed for wind, this again makes onshore wind generation in particular more attractive. This displaces offshore wind and PV in equal measure and also reduces the need for storage. However, lithium-ion batteries are also more affected here than Carnot batteries. Fig. 15. Annual electricity generation across Europe (VRE scenarios). F. Nitsch et al. F. Nitsch 3 Publications 55 Journal of Energy Storage 85 (2024) 110959 14 Fig. 16. Discharged energy across Germany (VRE scenarios, without methanizer). A focused look at Germany reveals some further effects (Fig. 16). For example, the very low contribution of lithium-ion batteries in the Base case is significantly increased by higher costs for gas generation and reconversion, and Carnot batteries also enter the system to a very small extent. In the opposite case of lower costs, however, there is no longer any room for batteries. Lower PV costs significantly increase the use of pumped hydro storage, but battery storage is no longer part of the system. In the case of reduced wind power costs, batteries are again not part of the optimal solution, and the use of pumped storage is also reduced. At first glance, the result for the case of reduced costs for wind and photovoltaics is surprising. This leads to an even greater increase in the use of electricity storage than a cost reduction for photovoltaics alone. In addition, not only is the use of lithium-ion batteries increased here, but Carnot batteries are also used. This results from the increased use of photovoltaics and onshore wind, which are supplemented by different storage systems. In contrast, the use of offshore wind and flexible CHP plants is reduced (Fig. 17). Fig. 17. Annual electricity generation in Germany (VRE scenarios). Appendix C. Spatial distribution of Carnot batteries in Germany In the case of Germany, the techno-economic targets for Carnot batteries need to be quite ambitious in order to arrive at relevant capacities. This effect is more prominent due to the sectoral representation with a detailed heating and gas sector. Both sectors allow for flexible demand via heat pumps, electric boilers and electrolysis, which in turn reduce the overall storage demand. In addition, due to the optimistic assumptions in the chosen techno-economic configuration lithium-ion batteries are almost completely pushed out of the system. As a result, in the Base case (Fig. 18a) Carnot batteries are only expanded in the Southern regions of Germany while with less flexible demand for electrolysis (Fig. 18b) and prevention of grid F. Nitsch et al. F. Nitsch 3 Publications 56 Journal of Energy Storage 85 (2024) 110959 15 expansion (Fig. 18c) the overall demand for storage technologies increases and Carnot batteries are expanded in more model regions. Fig. 18. Spatial distribution of annually provided energy from Carnot batteries for the three scenarios base (a), low flexibility electrolysis (b) and no additional grid expansion (c). Values are derived from the techno-economical configuration of 65 % round-trip efficiency, 20 € /kWh storage specific CAPEX and 150 € /kW power specific CAPEX. References [1] F. Nitsch, M. Wetzel, Profitability of power-to-heat-to-power storages in scenarios with high shares of renewable energy, Energy Proceedings 28 (2022), https://doi. org/10.46855/energy-proceedings-10258. [2] F.J. de Sisternes, J.D. Jenkins, A. Botterud, The value of energy storage in decarbonizing the electricity sector, Appl. Energy 175 (2016) 368–379, https:// doi.org/10.1016/j.apenergy.2016.05.014. [3] W.A. Braff, J.M. Mueller, J.E. Trancik, Value of storage technologies for wind and solar energy, Nat. Clim. Chang. 6 (10) (2016) 964–969, https://doi.org/10.1038/ nclimate3045. [4] S. Koohi-Fayegh, M.A. Rosen, A review of energy storage types, applications and recent developments, J. Energy Storage 27 (2020) 101047, https://doi.org/ 10.1016/j.est.2019.101047. [5] A.B. Gallo, J.R. Sim˜ oes-Moreira, H. Costa, M.M. Santos, E. Moutinho dos Santos, Energy storage in the energy transition context: a technology review, Renew. Sustain. Energy Rev. 65 (2016) 800–822, https://doi.org/10.1016/j.rser.2016.07. 028. [6] I. Sarbu, C. Sebarchievici, A comprehensive review of thermal energy storage, Sustainability 10 (1) (2018) 191, https://doi.org/10.3390/su10010191. [7] B. Lehner, G. Czisch, S. Vassolo, The impact of global change on the hydropower potential of Europe: a model-based analysis, Energy Policy 33 (7) (2005) 839–855, https://doi.org/10.1016/j.enpol.2003.10.018. [8] A. Aghahosseini, C. Breyer, Assessment of geological resource potential for compressed air energy storage in global electricity supply, Energ. Conver. Manage. 169 (2018) 161–173, https://doi.org/10.1016/j.enconman.2018.05.058. [9] IEA, Global EV Outlook 2020. Paris [Online]. Available: https://www.iea.org/r eports/global-ev-outlook-2020, 2020. [10] P.A. Christensen, et al., Risk management over the life cycle of lithium-ion batteries in electric vehicles, Renew. Sustain. Energy Rev. 148 (2021) 111240, https://doi.org/10.1016/j.rser.2021.111240. [11] C. Minke, U. Kunz, T. Turek, Techno-economic assessment of novel vanadium redox flow batteries with large-area cells, J. Power Sources 361 (2017) 105–114, https://doi.org/10.1016/j.jpowsour.2017.06.066. [12] M. Jafari, A. Botterud, A. Sakti, Decarbonizing power systems: a critical review of the role of energy storage, Renew. Sustain. Energy Rev. 158 (2022) 112077, https://doi.org/10.1016/j.rser.2022.112077. [13] W.-D. Steinmann, Thermo-mechanical concepts for bulk energy storage, Renew. Sustain. Energy Rev. 75 (2017) 205–219, https://doi.org/10.1016/j.rser.2016. 10.065. [14] V. Novotny, V. Basta, P. Smola, J. Spale, Review of Carnot battery technology commercial development, Energies 15 (2) (2022) 647, https://doi.org/10.3390/ en15020647. [15] A. Paul, F. Holy, M. Textor, S. Lechner, High temperature sensible thermal energy storage as a crucial element of Carnot batteries: overall classification and technical review based on parameters and key figures, J. Energy Storage 56 (2022) 106015, https://doi.org/10.1016/j.est.2022.106015. [16] M. Moser, H.-C. Gils, G. Pivaro, A sensitivity analysis on large-scale electrical energy storage requirements in Europe under consideration of innovative storage technologies, J. Clean. Prod. 269 (2020) 122261, https://doi.org/10.1016/j. jclepro.2020.122261. [17] O. Dumont, G.F. Frate, A. Pillai, S. Lecompte, M. De Paepe, V. Lemort, Carnot battery technology: a state-of-the-art review, J. Energy Storage 32 (2020) 101756, https://doi.org/10.1016/j.est.2020.101756. [18] A. Basta, V. Basta, J. Spale, T. Dlouhy, V. Novotny, Conversion of combined heat and power coal-fired plants to Carnot batteries - prospective sites for early gridscale applications, J. Energy Storage 55 (2022) 105548, https://doi.org/10.1016/j. est. 2022.105548. [19] P. Sorknæs, J.Z. Thellufsen, K. Knobloch, K. Engelbrecht, M. Yuan, Economic potentials of Carnot batteries in 100% renewable energy systems, Energy 282 (2023) 128837, https://doi.org/10.1016/j.energy.2023.128837. [20] Y. Zhang, L. Xu, J. Li, L. Zhang, Z. Yuan, Technical and economic evaluation, comparison and optimization of a Carnot battery with two different layouts, J. Energy Storage 55 (2022) 105583, https://doi.org/10.1016/j.est.2022.105583. [21] R. Fan, H. Xi, Energy, exergy, economic (3E) analysis, optimization and comparison of different Carnot battery systems for energy storage, Energ. Conver. Manage. 252 (2022) 115037, https://doi.org/10.1016/j.enconman.2021.115037. [22] S. Liu, H. Bai, P. Jiang, Q. Xu, M. Taghavi, Economic, energy and exergy assessments of a Carnot battery storage system: comparison between with and without the use of the regenerators, J. Energy Storage 50 (2022) 104577, https:// doi.org/10.1016/j.est.2022.104577. [23] T. Liang, et al., Key components for Carnot battery: technology review, technical barriers and selection criteria, Renew. Sustain. Energy Rev. 163 (2022) 112478, https://doi.org/10.1016/j.rser.2022.112478. [24] D. Luca de Tena, T. Pregger, Impact of electric vehicles on a future renewable energy - based power system in Europe with a focus on Germany, Int. J. Energy Res. 0 (0) (2018), https://doi.org/10.1002/er.4056. [25] H.C. Gils, Balancing of intermittent renewable power generation by demand response and thermal energy storage, Universit¨ at Stuttgart (2015), https://doi.org/ 10.18419/opus-6888. [26] T. Brown, D. Schlachtberger, A. Kies, S. Schramm, M. Greiner, Synergies of sector coupling and transmission reinforcement in a cost-optimised, highly renewable European energy system, Energy 160 (2018) 720–739, https://doi.org/10.1016/j. energy.2018.06.222. [27] H.C. Gils, Y. Scholz, T. Pregger, D. Luca de Tena, D. Heide, Integrated modelling of variable renewable energy-based power supply in Europe, Energy 123 (2017) 173–188, https://doi.org/10.1016/j.energy.2017.01.115. [28] C. Schimeczek, et al., AMIRIS: agent-based market model for the investigation of renewable and integrated energy systems, JOSS 8 (84) (2023) 5041, https://doi. org/10.21105/joss.05041. [29] F. Nitsch, C. Schimeczek, M. Wetzel, J. Buschmann, K. von Krbek, iog2x - Input/ Output Processing of GDX Files, 2023, https://doi.org/10.5281/zenodo.8283336. [30] B. Fuchs, J. Vesper, F. Nitsch, N. Wulff, ioProc — a Light-Weight Workflow Manager in Python, 2020, https://doi.org/10.5281/zenodo.4010275. [31] Y. Scholz, Renewable Energy Based Electricity Supply at Low Costs: Development of the REMix Model and Application for Europe, Universit¨ at Stuttgart, 2012, https://doi.org/10.18419/opus-2015. [32] K.-K. Cao, Modeling Energy Scenarios with Power-Flow Constraints: Transparency, Challenges and System Adequacy, 2020. [33] H.C. Gils, H. Gardian, J. Schmugge, Interaction of hydrogen infrastructures with other sector coupling options towards a zero-emission energy system in Germany, Renew. Energy 180 (2021) 140–156, https://doi.org/10.1016/j. renene.2021.08.016. [34] D. Luca de Tena, Large Scale Renewable Power Integration with Electric Vehicles, Universit¨ at Stuttgart, 2014 [Online]. Available: http://elib.uni-stuttgart.de/opus /volltexte2014/9407. [35] S. Sasanpour, K.-K. Cao, H.C. Gils, P. Jochem, Strategic policy targets and the contribution of hydrogen in a 100% renewable European power system, Energy Rep. 7 (2021) 4595–4608, https://doi.org/10.1016/j.egyr.2021.07.005. [36] H.C. Gils, S. Simon, Carbon neutral archipelago −100% renewable energy supply for the Canary Islands, Appl. Energy 188 (2017) 342–355, https://doi.org/ 10.1016/j. apenergy.2016.12.023. F. Nitsch et al. F. Nitsch 3 Publications 57 Journal of Energy Storage 85 (2024) 110959 16 [37] M. Wetzel, H.C. Gils, V. Bertsch, Green energy carriers and energy sovereignty in a climate neutral European energy system, Renew. Energy 210 (2023) 591–603, https://doi.org/10.1016/j.renene.2023.04.015. [38] H.C. Gils, et al., Model-related outcome differences in power system models with sector coupling—quantification and drivers, Renew. Sustain. Energy Rev. 159 (2022) 112177, https://doi.org/10.1016/j.rser.2022.112177. [39] H.C. Gils, et al., REMix Model Input Data for the THG95/GHG95 Scenario Analysed within the MuSeKo Project, 2021. [40] A. Vecchi, et al., Carnot battery development: a review on system performance, applications and commercial state-of-the-art, J. Energy Storage 55 (2022) 105782, https://doi.org/10.1016/j.est.2022.105782. [41] O. Dumont, V. Lemort, Mapping of performance of pumped thermal energy storage (Carnot battery) using waste heat recovery, Energy 211 (2020) 118963, https:// doi.org/10.1016/j.energy.2020.118963. [42] C. Schimeczek, et al., FAME-core: an open framework for distributed agent-based modelling of energy systems, JOSS 8 (84) (2023) 5087, https://doi.org/10.21105/ joss. 05087. [43] F. Nitsch, C. Schimeczek, U. Frey, and B. Fuchs, "FAME-Io: configuration tools for complex agent-based simulations," JOSS, vol. 8, no. 84, p. 4958, 2023, doi: 10.21105/ joss.04958. [44] S. Sarfarazi, M. Deissenroth-Uhrig, V. Bertsch, Aggregation of households in community energy systems: an analysis from actors’ and market perspectives, Energies 13 (19) (2020) 5154. [45] F. Nitsch, M. Deissenroth-Uhrig, C. Schimeczek, V. Bertsch, Economic evaluation of battery storage systems bidding on day-ahead and automatic frequency restoration reserves markets, Appl. Energy 298 (2021) 117267, https://doi.org/10.1016/j. apenergy. 2021.117267. [46] U.J. Frey, M. Klein, K. Nienhaus, C. Schimeczek, Self-reinforcing electricity price dynamics under the variable market premium scheme, Energies 13 (20) (2020), https://doi.org/10.3390/en13205350. [47] M. Reeg, AMIRIS-ein agentenbasiertes Simulationsmodell zur akteursspezifischen Analyse techno-¨ okonomischer und soziotechnischer Effekte bei der Strommarktintegration und Refinanzierung erneuerbarer Energien, Deutsches Zentrum für Luft-und Raumfahrt, Institut für Technische Thermodynamik, 2019. [48] F. Nitsch, C. Schimeczek, S. Wehrle, Back-testing the agent-based model AMIRIS for the Austrian day-ahead electricity market, 2021 [Online]. Available: https:// zenodo.org/record/5726738. [49] K. Nienhaus, et al., AMIRIS Examples [Online]. Available: https://zenodo.org/do i/10.5281/zenodo.7789049, 2023. [50] K. Nienhaus, et al., AMIRIS. Agent-based Market model for the Investigation of Renewable and Integrated energy Systems. https://gitlab.com/dlr-ve/esy/amir is/amiris, GitLab, . [51] ENTSO-E, Ten-Year Network Development Plan 2016 Executive Report, ENTSO-E, 2016 [Online]. Available: https://tyndp.entsoe.eu/2016/exec-report/. [52] K. Bruninx, et al., E-highway2050: D2. 1 Data sets of scenarios for 2050 [Online]. Available: http://www.e-highway2050.eu/fileadmin/documents/Results/e-High way_database_per_country-08022016.xlsx. [53] H. Lange, A. Klose, W. Lippmann, L. Urbas, Technical evaluation of the flexibility of water electrolysis systems to increase energy flexibility: a review, Int. J. Hydrogen Energy 48 (42) (2023) 15771–15783, https://doi.org/10.1016/j. ijhydene. 2023.01.044. [54] A. Dietrich, F. Dammel, P. Stephan, Exergoeconomic analysis of a pumped heat electricity storage system based on a Joule/Brayton cycle, Energy Sci. Eng. 9 (5) (2021) 645–660, https://doi.org/10.1002/ese3.850. [55] D. Scharrer, P. Bazan, M. Pruckner, R. German, Simulation and analysis of a Carnot battery consisting of a reversible heat pump/organic Rankine cycle for a domestic application in a community with varying number of houses, Energy 261 (2022) 125166, https://doi.org/10.1016/j.energy.2022.125166. [56] J.L.D. La Quintana, A. Sebasti´ an, R. Abbas, Economic assessment of installed CSP plants as Carnot Batteries: Spanish grid market case study, in: SOLARPACES 2020: 26th International Conference on Concentrating Solar Power and Chemical Energy Systems, Freiburg, Germany, 2022, p. 50002. [57] R. Tassenoy, K. Couvreur, W. Beyne, M. De Paepe, S. Lecompte, Techno-economic assessment of Carnot batteries for load-shifting of solar PV production of an office building, Renew. Energy 199 (2022) 1133–1144, https://doi.org/10.1016/j. renene.2022.09.039. [58] J. Schaffert, et al., Integrating system and operator perspectives for the evaluation of power-to-gas plants in the future German energy system, Energies 15 (3) (2022) 1174, https://doi.org/10.3390/en15031174. [59] Danish Energy Agency, Technology data - energy plants for electricity and district heating generation: Technical Report [Online]. Available: https://ens.dk/en/o ur-services/projections-andmodels/technology-data, 2023. Glossary ABM: Agent-based electricity market model AMIRIS: ABM developed at the German Aerospace Center CAPEX: Capital expenditure E2P: Energy-to-power (ratio) REMix: Framework for optimizing energy system models VRE: Variable renewable energy F. Nitsch et al. F. Nitsch 3 Publications 58 MAE =1 T∑ T t=1 |at− pt|(4) where a and  p represent the actual and forecasted prices, respectively, with a total length of T. We chose an absolute error metric, such as MAE, given the potential for target values (electricity prices) to be (close to) zero or even negative, making the use of mean absolute percentage errors (MAPE) problematic. Additionally, we calculated root-meansquared errors (RMSE) as RMSE =√{1 T∑T t=1(at− pt)2}(5) 4. Results We begin by examining the results of the AMIRIS scenarios with different shares of flexibility options followed by the results of forecasting accuracy in scenarios of RE expansion. 4.1. Different shares of flexibility options Prior to presenting the results of the forecasting methods in Section 4.1.2, we analyze the simulation runs of the four scenarios. These results offer valuable background information that aids in interpreting the forecasting accuracy. 4.1.1. Impact on market dynamics Fig. 2 shows day-ahead electricity prices as simulated by AMIRIS over a 168-hour period, with each line representing one of the four scenarios (essentially varying the amount of flexibility options available). Increasing storage capacity generally has a dampening effect on electricity prices. In particular, price peaks can be flattened by discharging storage, while valleys can be raised by charging storage. It is important to note that the storage operator aims to maximize its profits, while taking into account its impact on electricity prices when optimizing its bidding schedule. As a result, there are certain time periods when all four curves are aligned, indicating that during these hours the storage operator either has no significant impact on the resulting electricity prices, or is simply inactive. The statistics in Table 2 provide an understanding of the variability of electricity prices under different assumptions of storage installations, highlighting the impact of flexibility on pricing dynamics. In “No Flex”, there is a relatively wide dispersion of prices, as indicated by the standard deviation of 15.27 EUR/MWh. The minimum price 2 is −63.51 EUR/MWh, while the maximum is 116.83 EUR/MW. Moving to “Little Flex”, the mean price increases slightly to 38.49 EUR/MWh, accompanied by a lower standard deviation of 13.75 EUR/MWh, indicating a narrower spread of prices compared to “No Flex”. In “Mid Flex”, the mean price rises further to 38.80 EUR/MWh, with a continued decrease in the standard deviation to 12.07 EUR/MWh, suggesting the previously described price dampening effect. Finally, in “High Flex”, the mean price reaches 38.96 EUR/MWh, accompanied by the lowest standard deviation of 11.17 EUR/MWh among all scenarios. The results from the year 2018 demonstrate similar overall trends as found for the year 2019. However, differences arise in the weighted mean electricity prices amounting to 43.50 EUR/MWh in 2018 and 38.50 EUR/MWh in 2019. 4.1.2. Electricity price forecasting accuracy The two ML methods N-BEATS and TFT are trained on 2018 and tested on 2019. For the evaluation of the forecasting accuracy, a full year is chunked in roughly 500 samples, each with 168 hours of past covariate data and 24 hours to forecast. Forecasted values are then tested against actual values from the simulation. Table 3 lists MAE of all forecasting methods in the four scenarios. The MAE provide insights into the forecasting capabilities of each method within different shares of flexibility options. Notably, the TFT trained with future covariates (expected load and RE generation) shows the lowest MAE values, suggesting its superior accuracy in forecasting electricity prices across a spectrum of scenarios. The benchmark methods are performing worse in every scenario, but are much cheaper to apply, since they do not require to train a model. Further, the presented results clearly demonstrate a consistent trend of improved accuracy across the scenarios of “No Flex” to “High Flex”. We conclude that this effect is likely due to the price flattening effect of increasing market impact of flexibility options, see also Fig. 2. Fig. 3 displays an exemplary forecast made by the TFT model in “Mid Flex”. Overall, the forecast aligns well with the actual price dynamics. Nevertheless, there is a slight deviation as the model fails to accurately forecast the first valley underestimating actual values. This can likely be attributed to price dynamics caused by charging actions of flexibility options. Forecasting errors also exhibit a temporal dependency that correlates with the load pattern. Consequently, accuracy tends to be best at night, with errors peaking during the day (Nitsch and Schimeczek, 2023). In the context of the German case study presented, local weather effects, including short-term fluctuations in renewable energy generation, generally have a limited impact on the day-ahead market zone and are therefore not of great influence to this particular forecasting procedure. 4.2. Forecasting accuracy in future energy scenarios As elaborated in the Introduction in Section 1, the energy transition will bring very different market dynamics compared to historical observations. Besides the expanding flexibility potential, as presented in Section 4.1, we expect considerable impacts by the expansion of RE leading to novel price dynamics of electricity markets. Therefore, we investigate the effects on forecasting accuracy in such scenarios. For this we have created unique training and testing data, as described in Section 3.2. Due to high computational costs of ML training, we have limited this analysis to TFT, the best performing method so far. Fig. 4 shows the results of different train-test splits in two model configurations (without future covariates and with future covariates). All six models are trained and evaluated independently. The left column shows the available training data to the TFT model and their weighted mean average prices Fig. 2. Price dampening effect of different flexibility capacity in the four scenarios on simulated electricity prices over a one-week period in November 2019. 2 Periods of high inflexible generation and low demand can lead to negative prices in electricity markets. F. Nitsch et al. Energy Reports 12 (2024) 5268–5279 5272 F. Nitsch 3 Publications 65 of the scenario marked with ‘x’. Values in between are interpolated by a cubic method. It is evident that as RE capacity increases, electricity prices tend to decrease – a trend consistently observed across all three rows representing different numbers of scenarios and train-test splits (10 scenarios, 30 scenarios, and 90 scenarios). The middle and right column show the forecasting accuracy evaluated as MAE over all errors of the forecasted electricity prices against the actual electricity prices as calculated by AMIRIS. In the middle column, the TFT relies solely on past covariates, while in the right column, the TFT also incorporates future covariates, such as calendar information, expected load, and RE generation. Notably, we observe a robust results with MAE values ranging from 2.25 to 3.25 EUR/MWh when at least 30 scenarios are employed as training data (middle and bottom rows). However, when restricting training data to only 10 scenarios (top row), forecasting accuracy deteriorates significantly, with MAE doubling, when assessing scenarios that fall well outside the range of known training data. This strongly suggests that the selection of training data is important for the performance of the model. Moreover, the TFT model equipped with future covariates (right column) consistently outperforms the version relying solely on past covariates (middle). Errors are reduced by approximately an order of magnitude, which holds promise for applications in energy system models, particularly ABM. Moreover, in the bottom row, where the train-test split is 75 % and 25 %, the results are similar compared to the middle row, where the split is reversed at 25 % and 75 %. Additionally, we can observe that MAE generally exhibits a downward trend as onshore wind capacity increases, except for the segment of high wind power capacity, which lacked sufficient training data, as indicated in the top row of plots in Fig. 4. A similar trend is also evident when considering different error metrics like RMSE. The histograms in Fig. 5 illustrate the distribution of errors for training with 30 and 90 scenarios. It can be observed that an increased number of training scenarios leads to a superior fit when MAE is employed as the error metric. However, the addition of future covariates – a common practice in such forecasting problems (Ozyegen et al., 2022) – improves the accuracy in our analysis even more than the quantity of available training data. Specifically, with such covariate data, MAE remain consistently below 1.40 EUR/MWh. 5. Discussion As suggested in Haugen et al. (2024), the formation of electricity prices in energy market models represents a significant factor influencing the analysis of actors’ behaviour, ranging from the operation of flexibility options to investment decisions. The use of simulated and synthetic data as a complement to historical data is an attractive approach, particularly given its current deployment in the context of creating energy generation and load profiles (Mayer et al., 2023). Consequently, the presented methodology in this paper contributes valuable insights to the currently limited field of such day-ahead electricity price forecasting in high RE penetration scenarios. However, given the inherent complexity and non-linearity of energy markets (Castilho Braz et al., 2024), it is essential to consider that our general conclusions should not be interpreted as individual predictions on market results. Rather, they should be regarded as projections contributing towards a better understanding of potential market dynamics in systems with high shares of RE. Despite the presentation of a comprehensive range of potential scenarios, uncertainty remains to scenario definition and model formulation. Future research should investigate Table 2 Descriptive statistics on simulated electricity prices in the four scenarios. Year 2018 2019 Scenario No Flex Little Flex Mid Flex High Flex No Flex Little Flex Mid Flex High Flex Metric Std. dev. 15.48 13.56 11.95 11.07 15.27 13.75 12.07 11.17 Minimum −35.21 −22.11 −17.18 0.00 −63.51 −52.49 −37.73 0.00 Maximum 115.96 103.38 95.16 94.78 116.83 102.15 96.33 96.33 Table 3 Mean absolute error (MAE) in EUR/MWh of forecasts in four different scenarios. Scenario No Flex Little Flex Mid Flex High Flex Metric Naïve t 1 (1) 9.29 7.78 6.76 6.45 Naïve t 24 (2) 8.57 7.54 6.27 5.91 Exponential Smoothing (3) 8.06 6.70 5.73 5.46 N-BEATS (Oreshkin et al., 2019) 7.15 6.24 5.38 5.12 TFT (Lim et al., 2021) 4.11 3.90 3.20 3.26 TFT with future covariates (Lim et al., 2021) 3.12 3.45 3.26 2.86 Fig. 3. Exemplary forecasted prices for the next 24 hours (green) by the TFT model in “Mid Flex” with 168 hours of past covariates plotted against actual prices (black). F. Nitsch et al. Energy Reports 12 (2024) 5268–5279 5273 F. Nitsch 3 Publications 66 the influence of changes in market and policy design, unforeseen events with significant impact on energy markets, and the manner in which market actors respond to forecast uncertainty. In addition, we wish to highlight the following limitations. In the realm of analyzing various flexibility option shares in Section 4.1, it is important to acknowledge a potential limitation related to model training. A possible enhancement lies in the inclusion of a more diverse set of training data to refine the robustness of our models. The analysis of RE expansion in Section 4.2 underscores the considerable impact of the training data selection on results. Future analyses should therefore diligently consider this important aspect. Similarly, within the analysis of RE expansion scenarios in Section 4.2, a notable limitation lies in the missing variability of weather conditions. Periods of low wind and solar generation may substantially impact results, particularly as RE capacity increases. The use of NN with future covariates (as demonstrated in Fig. 4) might mitigate this impact since the network is aware of the short-term residual load. However, without these future covariates, variations in weather years could exert a greater influence on the results. To ensure that our results are easily transferable and understandable, our presented scenarios assume no variation in parameters aside from wind and PV capacity. As already described in Section 3.2, this assumption does not fully capture the real-world dynamics of the energy transition missing the evolution of flexible demand and generation capacity. It is evident that additional markets, which are currently under discussion but not yet implemented, such as capacity or flexibility markets, would influence market dynamics and necessitate further analysis. However, an examination of these points is beyond the scope of the present manuscript. Beyond these specific limitations, broader considerations should be mentioned. The computational resources and training data allocation significantly affect the time required for training NN. While the initial effort to train models and optimize hyperparameters is substantial, transitioning to the utilization of pre-trained models with optional finetuning in production settings could significantly alleviate this workload. When experimenting with a wider array of input features, the explainable feature of TFT could help identify the most influential factors governing forecast accuracy. Additionally, the incorporation of TFT’s capability to provide probabilistic forecasts holds the potential to Fig. 4. Simulation of PV and onshore wind expansion scenarios (marked by white ’x’ markers) using AMIRIS. Values in between are interpolated using a cubic method. The left column represents the training data, while the other two columns illustrate forecasting accuracy in terms of Mean Absolute Error (MAE) in EUR/ MWh. The middle column shows the accuracy based on past covariates alone, whereas the right column includes additional future covariates to the forecasting procedure. Note: Scenarios are considered as parameter variations and shall not be interpreted as definitive and complete future electricity systems, see also Section 3.2. F. Nitsch et al. Energy Reports 12 (2024) 5268–5279 5274 F. Nitsch 3 Publications 67 broaden the applications within energy market simulation models. In terms of our findings, they align with existing literature as follows: Lago et al. (2021) conducted an extensive review of day-ahead electricity price forecasting, also concluding that deep neural NN, such as TFT, tend to outperform Lasso Estimated AutoRegressive methods, albeit with increased computational costs. Fraunholz et al. (2021) found that NN outperform regression and naïve benchmarks when applied in an ABM. However, the choice of specific architecture significantly influences the results of the ABM, underscoring the need for a careful assessment. Trebbien et al. (2023) presented an analysis of day-ahead electricity prices from 2017–2020, identifying load, wind, and solar generation as key features for an explainable ML model, aligning with our findings regarding input data selection. While our study focuses on TFT networks in a similar domain, it is worth noting that they are often employed in forecasting load (Nazir et al., 2023) or renewable energy generation (L´ opez Santos et al., 2022). In contrast to the original N-BEATS architecture (Oreshkin et al., 2019), which does not allow future covariates, the N-BEATSx (Olivares et al., 2023) offers this extension. Numerical test results show MAE of around 3.30 EUR/MWh on historical German market data. Azam and Younis (2021) conduct load and price forecasting using a novel hybrid deep learning approach demonstrating achieving MAE of around 5.20 USD/MWh on the ISO New England energy market in 2018 and 2019. In Ziel and Weron (2018), twelve distinct historical datasets of day-ahead electricity prices are evaluated, revealing a MAE in the German market zone of approximately 5 EUR/MWh. Fraunholz et al. (2021) perform a scenario analysis of ten interconnected market zones in Europe from 2020 to 2050 with MAPE forecasting errors between 0.10 and 0.39. 6. Conclusion The findings of our study demonstrate a powerful approach that combines agent-based electricity market simulation and time series forecasting based on machine learning to provide forecasts for energy transition scenarios. Past and present market data, which are widely used in forecasting studies today, do not account for the novel price dynamics of future highly renewable electricity markets. In contrast, we explicitly incorporate foreseeable changes in energy systems resulting from the ongoing energy transition. In particular, we investigate energy transition scenarios with significant expansion of flexibility options and renewable energies, which are used to train and test different forecasting methods. We then use an open state-of-the-art agent-based electricity market model and open data to generate market results in these scenarios that differ significantly from today’s energy system. We then assess the accuracy of different electricity price forecasting methods in the context of these widely varying scenarios. In our assessment, comprehensive machine learning methods, namely Temporal Fusion Transformers, demonstrate superior forecasting accuracy for future electricity markets compared to naïve benchmarking methods. Mean absolute errors decrease by approximately one order of magnitude when future covariates are accessible and understandable to the model. In addition to the demonstrated precision, even in environments characterized by significant change, the examined methodologies offer several key advantages over conventional forecasting techniques. Some machine learning-based methods, including Transformers, are capable of Fig. 5. Distribution of mean absolute error (MAE) in EUR/MWh for different training sets and TFT configurations. Note: Different scaling of x-axis for runs with future covariates. F. Nitsch et al. Energy Reports 12 (2024) 5268–5279 5275 F. Nitsch 3 Publications 68 handling disparate input data configurations, thereby facilitating their adaptation to evolving settings. Moreover, scaling is readily achievable from simple proof-of-concepts to comprehensive modelling suites. In order to apply our results to other electricity market simulations, modelers need to apply their domain knowledge when defining training data and selecting input features. As machine learning-based methods can be computationally expensive, adequate resources are required, at least during the initial training stage. Our results are relevant not only to agent-based electricity market modelling but also to the broader field of electricity price forecasting. The presentation of quantitative results on forecasting accuracy contributes valuable insights to the general understanding of modeling electricity markets affected by the energy transition. Furthermore, they can be employed to supplement existing assessments of investment decisions within the industrial sector. From a technical perspective, modular, open, and comprehensive software packages facilitate the transferability of our approach to other applications and more in-depth analyses. Future research may address broadening the scenario space, with specific attention to the incorporation of diverse storage agents, varying technological considerations, the influence of potential market powers, and the impact of different agents’ operational strategies. It would also be valuable to investigate the uncertainty of future electricity market scenarios in terms of market design and agent behaviour. Furthermore, the presented forecasting technique could also be applied to additional markets, such as Intraday markets. This would facilitate a more comprehensive analysis of the interplay between multiple markets. CRediT authorship contribution statement Felix Nitsch: Writing – original draft, Visualization, Validation, Software, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Christoph Schimeczek: Writing – review & editing, Supervision, Funding acquisition. Valentin Bertsch: Writing – review & editing, Supervision, Conceptualization. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements Felix Nitsch and Christoph Schimeczek would like to express their gratitude to the German Federal Ministry of Education and Research (BMBF) for their financial support of the FEAT project (grant number 01IS22073B). The authors would like to thank Kristina Nienhaus, Ulrich Frey, A. Achraf El Ghazi, Benjamin Fuchs, and the colleagues from the Department of Energy Systems Analysis at the Institute of Networked Energy Systems, German Aerospace Center (DLR), for their invaluable feedback and insightful discussions on first drafts of the manuscript. Author agreement statement We the declare that this manuscript is original, has not been published before and is not currently being considered for publication elsewhere. We confirm that the manuscript has been read and approved by all named authors and that there are no other persons who satisfied the criteria for authorship but are not listed. We further confirm that the order of authors listed in the manuscript has been approved by all of us. We understand that the Corresponding Author, Felix Nitsch, is the sole contact for the Editorial process. He is responsible for communicating with the other authors about progress, submissions of revisions and final approval of proofs. Appendix F. Nitsch et al. Energy Reports 12 (2024) 5268–5279 5276 F. Nitsch 3 Publications 69 Fig. 6. Schematic overview of the agents and their connections in the agent-based electricity market model AMIRIS (Schimeczek et al., 2023a). Data Availability All code used to run this analysis is openly available in Schimeczek et al. (2023a), Nitsch et al. (2023b), Nitsch (2023). The data is based on Nienhaus et al. (2023). References Akhtar, S., Adeel, M., Iqbal, M., Namoun, A., Tufail, A., Kim, K.-H., 2023. Deep learning methods utilization in electric power systems. Energy Rep. 10, 2138–2151. Akiba T., Sano S., Yanase T., Ohta T., Koyama M. Optuna: A Next-generation Hyperparameter Optimization Framework; 2019. Alhendi, A., Al-Sumaiti, A.S., Marzband, M., Kumar, R., Diab, A.A.Z., 2023. Short-term load and price forecasting using artificial neural network with enhanced Markov chain for ISO New England. Energy Rep. 9, 4799–4815. Alkhayat, G., Mehmood, R., 2021. A review and taxonomy of wind and solar energy forecasting methods based on deep learning. Energy AI, 100060. Arslan Tuncar, E., Sa˘ glam, S¸., Oral, B., 2024. A review of short-term wind power generation forecasting methods in recent technological trends. Energy Rep. 12, 197–209. Aussel, D., Bendotti, P., Piˇ stˇ ek, M., 2017. Nash equilibrium in a pay-as-bid electricity market: Part 1-existence and characterization. Optimization 66 (6), 1013–1025. Azam, M.F., Younis, M.S., 2021. Multi-horizon electricity load and price forecasting using an interpretable multi-head self-attention and EEMD-Based framework. IEEE Access 9, 85918–85932. Bai, Z., 2024. Residential electricity prediction based on GA-LSTM modeling. Energy Rep. 11, 6223–6232. Bale, C.S., Varga, L., Foxon, T.J., 2015. Energy and complexity: new ways forward. Appl. Energy 138, 150–159. Barazza, E., Strachan, N., 2020. The impact of heterogeneous market players with bounded-rationality on the electricity sector low-carbon transition. Energy Policy 138, 111274. Bashir, T., Haoyong, C., Tahir, M.F., Liqiang, Z., 2022. Short term electricity load forecasting using hybrid prophet-LSTM model optimized by BPNN. Energy Rep. 8, 1678–1686. Beltr´ an, S., Castro, A., Irizar, I., Naveran, G., Yeregui, I., 2022. Framework for collaborative intelligence in forecasting day-ahead electricity price. Appl. Energy 306, 118049. Beran P., Vogler A., Weber C. Multi-Day-Ahead Electricity Price Forecasting: A Comparison of fundamental, econometric and hybrid Models; 2021. Bill´ e, A.G., Gianfreda, A., Del Grosso, F., Ravazzolo, F., 2023. Forecasting electricity prices with expert, linear, and nonlinear models. Int. J. Forecast. 39 (2), 570–586. Camelo, do Nascimento, Lucio, H., Junior, P.S., Leal, Jo˜ ao Bosco Verçosa, Carvalho, P.C. M. de, Santos, dos, 2018. Daniel von Glehn. Innovative hybrid models for forecasting time series applied in wind generation based on the combination of time series models with artificial neural networks. Energy 151, 347–357. Castilho Braz, de, D.D., dos Santos, Paula, M.R., de, M.B.S., Da Silva Filho, D., Guarnier, E., Alípio, L.P., et al., 2024. Multi-source data ensemble for energy price trend forecasting. Eng. Appl. Artif. Intell. 133, 108125. Chappin, Emile J.L., de Vries, Laurens J., Richstein, Joern C., Bhagwat, Pradyumna, 2017. Kaveri Iychettira, Salman Khan. Simulating climate and energy policy with agent-based modelling: the energy modelling laboratory (EMLab). Environ. Model. Softw. 96, 421–431. Cheng, H.-Y., Yu, C.-C., Lin, C.-L., 2021. Day-ahead to week-ahead solar irradiance prediction using convolutional long short-term memory networks. Renew. Energy. Cruz, A., Mu˜ noz, A., Zamora, J.L., Espinola, R., 2011. The effect of wind generation and weekday on Spanish electricity spot price forecasting. Electr. Power Syst. Res. 81 (10), 1924–1935. Da Silva, D.G., Meneses, 2023. AAdM. Comparing Long Short-Term Memory (LSTM) and bidirectional LSTM deep neural networks for power consumption prediction. Energy Rep. 10, 3315–3334. Deissenroth-Uhrig, M., Klein, M., Nienhaus, K., Reeg, M., 2017. Assessing the plurality of actors and policy interactions: agent-based modelling of renewable energy market integration. Complexity 2017. Divya, K.C., Østergaard, J., 2009. Battery energy storage technology for power systems—an overview. Electr. Power Syst. Res. 79 (4), 511–520. European Commission. European Green Deal Delivering on our targets; 2021. F. Nitsch et al. Energy Reports 12 (2024) 5268–5279 5277 F. Nitsch 3 Publications 70 Fan, S., Liao, J.R., Yokoyama, R., Chen, L., Lee, W.-J., 2009. Forecasting the wind generation using a two-stage network based on meteorological information. IEEE Trans. Energy Convers. 24 (2), 474–482. Farmer, J.D., Hepburn, C., Mealy, P., Teytelboym, A., 2015. A third wave in the economics of climate change. Environ. Resour. Econ. 62 (2), 329–357. Fraunholz, C., Kraft, E., Keles, D., Fichtner, W., 2021. Advanced price forecasting in agent-based electricity market simulation. Appl. Energy 290, 116688. Frey, U.J., Klein, M., Nienhaus, K., Schimeczek, C., 2020. Self-reinforcing electricity price dynamics under the variable market premium scheme. Energies 13 (20). Guerci, E., Rastegar, M.A., Cincotti, S., 2010. Agent-based modeling and simulation of competitive wholesale electricity markets. Handbook of power systems II. Springer, pp. 241–286. Gunduz, S., Ugurlu, U., Oksuz, I., 2023. Transfer learning for electricity price forecasting. Sustain. Energy, Grids Netw. 34, 100996. Hansen, P., Liu, X., Morrison, G.M., 2019. Agent-based modelling and socio-technical energy transitions: A systematic literature review. Energy Res. Soc. Sci. 49, 41–52. Harder, N., Qussous, R., Weidlich, A., 2023. Fit for purpose: modeling wholesale electricity markets realistically with multi-agent deep reinforcement learning. Energy AI 14, 100295. Haugen, M., Blaisdell-Pijuan, P.L., Botterud, A., Levin, T., Zhou, Z., Belsnes, M., et al., 2024. Power market models for the clean energy transition: state of the art and future research needs. Appl. Energy 357, 122495. Heidarpanah, M., Hooshyaripor, F., Fazeli, M., 2023. Daily electricity price forecasting using artificial intelligence models in the Iranian electricity market. Energy 263, 126011. Holt, C.C., 2004. Forecasting seasonals and trends by exponentially weighted moving averages. Int. J. Forecast. 20 (1), 5–10. Huang, X., Li, Q., Tai, Y., Chen, Z., Zhang, J., Shi, J., et al., 2021. Hybrid deep neural model for hourly solar irradiance forecasting. Renew. Energy 171, 1041–1060. Hyndman R.J., Athanasopoulos G. Forecasting: principles and practice. 2nd ed. Melbourne; 2018. Jedrzejewski, A., Lago, J., Marcjasz, G., Weron, R., 2022. Electricity price forecasting: the dawn of machine learning. IEEE Power Energy Mag. 20 (3), 24–31. Jiang L., Hu G. A Review on Short-Term Electricity Price Forecasting Techniques for Energy Markets. In: 2018 15th International Conference on Control, Automation, Robotics and Vision (ICARCV): IEEE; 2018, p. 937–944. Jiang, P., Nie, Y., Wang, J., Huang, X., 2023. Multivariable short-term electricity price forecasting using artificial intelligence and multi-input multi-output scheme. Energy Econ. 117, 106471. Klein, M., Frey, U.J., Reeg, M., 2019. Models within models-agent-based modelling and simulation in energy systems analysis. J. Artif. Soc. Soc. Simul. 22 (4). Kontochristopoulos, Y., Michas, S., Kleanthis, N., Flamos, A., 2021. Investigating the market effects of increased RES penetration with BSAM: a wholesale electricity market simulator. Energy Rep. 7, 4905–4929. Kraan, O., Kramer, G.J., Nikolic, I., 2018. Investment in the future electricity system-an agent-based modelling approach. Energy 151, 569–580. Lago, J., Marcjasz, G., Schutter, B., de, Weron, R., 2021. Forecasting day-ahead electricity prices: a review of state-of-the-art algorithms, best practices and an openaccess benchmark. Appl. Energy 293, 116983. Ledmaoui, Y., El Maghraoui, A., El Aroussi, M., Saadane, R., Chebak, A., Chehri, A., 2023. Forecasting solar energy production: a comparative study of machine learning algorithms. Energy Rep. 10, 1004–1012. Lehna, M., Scheller, F., Herwartz, H., 2022. Forecasting day-ahead electricity prices: a comparison of time series and neural network models taking external regressors into account. Energy Econ. 106, 105742. Li, Shi, 2012. Agent-based modeling for trading wind power with uncertainty in the dayahead wholesale electricity markets of single-sided auctions. Appl. Energy 99, 13–22. Liberopoulos, G., Andrianesis, P., 2016. Critical review of pricing schemes in markets with non-convex costs. Oper. Res. 64 (1), 17–31. Lim, B., Arık, S.¨ O., Loeff, N., Pfister, T., 2021. Temporal fusion transformers for interpretable multi-horizon time series forecasting. Int. J. Forecast. L´ opez Santos, M., García-Santiago, X., Echevarría Camarero, F., Bl´ azquez Gil, G., Carrasco Ortega, P., 2022. Application of temporal fusion transformer for day-ahead PV power forecasting. Energies 15 (14), 5232. Makkonen, S., Lahdelma, R., 2006. Non-convex power plant modelling in energy optimisation. Eur. J. Oper. Res. 171 (3), 1113–1126. Martin, A., Müller, J.C., Pokutta, S., 2014. Strict linear prices in non-convex European day-ahead electricity markets. Optim. Methods Softw. 29 (1), 189–221. .Maurer, F., Nitsch, F., Kochems, J., Schimeczek, C., Sander, V., Lehnhoff, S., 2024. Know Your Tools - A Comparison of Two Open Agent-Based Energy Market Models. In: 2024 20th International Conference on the European Energy Market (EEM). 〈https://ieeexplore.ieee.org/document/10609021〉. IEEE, p. 1–8. Mayer, M.J., Bir´ o, B., Szücs, B., Asz´ odi, A., 2023. Probabilistic modeling of future electricity systems with high renewable energy penetration using machine learning. Appl. Energy 336, 120801. Meng, A., Wang, P., Zhai, G., Zeng, C., Chen, S., Yang, X., et al., 2022. Electricity price forecasting with high penetration of renewable energy using attention-based LSTM network trained by crisscross optimization. Energy 254, 124212. Mo, S., Wang, H., Li, B., Xue, Z., Fan, S., Liu, X., 2024. Powerformer: a temporal-based transformer model for wind power forecasting. Energy Rep. 11, 736–744. Müller, I.M., 2021. Feature selection for energy system modeling: identification of relevant time series information. Energy AI 4, 100057. Nazir, M.S., Alturise, F., Alshmrany, S., Nazir, H., Bilal, M., Abdalla, A.N., et al., 2020. Wind generation forecasting methods and proliferation of artificial neural network: a review of five years research trend. Sustainability 12 (9), 3778. Nazir, A., Shaikh, A.K., Shah, A.S., Khalil, A., 2023. Forecasting energy consumption demand of customers in smart grid using Temporal Fusion Transformer (TFT). Results Eng. 17, 100888. Nienhaus, K., Reeg, M., Roloff, N., Deissenroth-Uhrig, M., Klein, M., Schimeczek, C., et al., 2021. AMIRIS. Agent-based Market model for the Investigation of Renewable and Integrated energy Systems. GitLab. https://gitlab.com/dlr-ve/esy/amiris/amir is. Nienhaus, K., Schimeczek, C., Frey, U., Sperber, E., Sarfarazi, S., Nitsch, F., et al., 2023. AMIRIS Examples. 〈https://gitlab.com/dlr-ve/esy/amiris/examples〉. GitLab. Nitsch, F., 2023. focapy: Timeseries forecasting in Python. GitLab. https://gitlab.com/fo capy.. Nitsch, F., Deissenroth-Uhrig, M., Schimeczek, C., Bertsch, V., 2021a. Economic evaluation of battery storage systems bidding on day-ahead and automatic frequency restoration reserves markets. Appl. Energy 298, 117267. Nitsch, F., Frey, U., Schimeczek, C., 2023b. scengen: A Scenario Generator for the Open Electricity Market Model AMIRIS. 〈https://zenodo.org/records/8382789〉. Zenodo. Nitsch, F., Schimeczek, C., Wehrle, S., 2021b. Back-testing the agent-based model AMIRIS for the Austrian day-ahead electricity market. 〈https://zenodo.org/records /5726737〉. Zenodo. Nitsch, F., Schimeczek, C., Frey, U., Fuchs, B., 2023a. FAME-Io: Configuration tools for complex agent-based simulations. JOSS 8 (84), 4958. Nitsch, F., Schimeczek, C., 2023. Comparison of electricity price forecasting methods for use in agent-based energy system models. IEWT, Vienna. Nitsch, F., Wetzel, M., Gils, H.C., Nienhaus, K., 2024. The future role of Carnot batteries in Central Europe: combining energy system and market perspective. J. Energy Storage 85, 110959. Nowotarski, J., Weron, R., 2018. Recent advances in electricity price forecasting: a review of probabilistic forecasting. Renew. Sustain. Energy Rev. 81, 1548–1568. Nyangon, J., Akintunde, R., 2024. Principal component analysis of day-ahead electricity price forecasting in CAISO and its implications for highly integrated renewable energy markets. Wiley Interdiscip. Rev.: Energy Environ. 13 (1). Olivares, K.G., Challu, C., Marcjasz, G., Weron, R., Dubrawski, A., 2023. Neural basis expansion analysis with exogenous variables: Forecasting electricity prices with NBEATSx. Int. J. Forecast. 39 (2), 884–900. Oreshkin B.N., Carpov D., Chapados N., Bengio Y. N-BEATS: Neural basis expansion analysis for interpretable time series forecasting; 2019. Ozyegen, O., Ilic, I., Cevik, M., 2022. Evaluation of interpretability methods for multivariate time series forecasting. Appl. Intell. (Dordr., Neth. ) 52 (5), 4727–4743. Pape, C., Hagemann, S., Weber, C., 2016. Are fundamentals enough? Explaining price variations in the German day-ahead and intraday power market. Energy Econ. 54 (ement C)), 376–387. Petropoulos, F., Apiletti, D., Assimakopoulos, V., Babai, M.Z., Barrow, D.K., Ben Taieb, S., et al., 2022. Forecasting: theory and practice. Int. J. Forecast. 38 (3), 705–871. Pfenninger, S., Hawkes, A., Keirstead, J., 2014. Energy systems modeling for twenty-first century energy challenges. Renew. Sustain. Energy Rev. 33, 74–86. Qu, J., Qian, Z., Pei, Y., 2021. Day-ahead hourly photovoltaic power forecasting using attention-based CNN-LSTM neural network embedded with multiple relevant and target variables prediction pattern. Energy 232, 120996. Ragwitz, F., Genoese, M., M¨ ost, M., 2007. D. Agent-based simulation of electricity markets – a literature review. Sensfuß. Reeg, M., 2019. AMIRIS-ein agentenbasiertes Simulationsmodell zur akteursspezifischen Analyse techno-¨ okonomischer und soziotechnischer Effekte bei der Strommarktintegration und Refinanzierung erneuerbarer. Energien. Ringler, P., Keles, D., Fichtner, W., 2016. Agent-based modelling and simulation of smart electricity grids and markets – a literature review. Renew. Sustain. Energy Rev. 57, 205–215. Schimeczek, C., Deissenroth-Uhrig, M., Frey, U., Fuchs, B., Ghazi, A.A.E., Wetzel, M., et al., 2023b. FAME-core: an open framework for distributed agent-based modelling of energy systems. JOSS 8 (84), 5087. Schimeczek, C., Nienhaus, K., Frey, U., Sperber, E., Sarfarazi, S., Nitsch, F., et al., 2023a. AMIRIS: Agent-based Market model for the Investigation of Renewable and Integrated energy Systems. JOSS 8 (84), 5041. Sensfuß F. Assessment of the impact of renewable electricity generation on the German electricity sector - An agent-based simulation approach; 2008. Sewdien, V.N., Preece, R., Torres, J.R., Rakhshani, E., van der Meijden, M., 2020. Assessment of critical parameters for artificial neural networks based short-term wind generation forecasting. Renew. Energy 161, 878–892. Shimomura, M., Keeley, A.R., Matsumoto, K., Tanaka, K., Managi, S., 2024. Beyond the merit order effect: Impact of the rapid expansion of renewable energy on electricity market price. Renew. Sustain. Energy Rev. 189, 114037. SMARD - German electricity market data platform. 〈https://www.smard.de〉2020. Trebbien, J., Gorj˜ ao, L.R., Praktiknjo, A., Sch¨ afer, B., Witthaut, D., 2023. Understanding electricity prices beyond the merit order principle using explainable AI. Energy AI, 100250. Tschora, L., Pierre, E., Plantevit, M., Robardet, C., 2022. Electricity price forecasting on the day-ahead market using machine learning. Appl. Energy 313, 118752. Vale, Z., Pinto, T., Praca, I., Morais, H., 2011. MASCEM: electricity markets simulation with strategic agents. IEEE Intell. Syst. 26 (2), 9–17. Walter, V., Wagner, A., 2024. Probabilistic simulation of electricity price scenarios using Conditional Generative Adversarial Networks. Energy AI 18, 100422. Weidlich, A., Veit, D., 2008. A critical survey of agent-based wholesale electricity market models. Energy Econ. 30 (4), 1728–1759. Weron, R., 2014. Electricity price forecasting: a review of the state-of-the-art with a look into the future. Int. J. Forecast. 30 (4), 1030–1081. F. Nitsch et al. Energy Reports 12 (2024) 5268–5279 5278 F. Nitsch 3 Publications 71 Winters, P.R., 1960. Forecasting sales by exponentially weighted moving averages. Manag. Sci. 6 (3), 324–342. Xiong, X., Qing, G., 2023. A hybrid day-ahead electricity price forecasting framework based on time series. Energy 264, 126099. Zakeri, B., Staffell, I., Dodds, P.E., Grubb, M., Ekins, P., J¨ a¨ askel¨ ainen, J., et al., 2023. The role of natural gas in setting electricity prices in Europe. Energy Rep. 10, 2778–2792. Zhen, H., Niu, D., Wang, K., Shi, Y., Ji, Z., Xu, X., 2021. Photovoltaic power forecasting based on GA improved Bi-LSTM in microgrid without meteorological information. Energy 231, 120908. Ziel, F., Weron, R., 2018. Day-ahead electricity price forecasting with high-dimensional structures: univariate vs. multivariate modeling frameworks. Energy Econ. 70, 396–420. F. Nitsch et al. Energy Reports 12 (2024) 5268–5279 5279 F. Nitsch 3 Publications 72 F. Nitsch 3.4 Paper IV: Profitability of competing flexibility options in renewable-dominated energy markets: Combining agent-based and machine learning approaches Authors: Felix Nitsch, Christoph Schimeczek, Valentin Bertsch Corresponding Author: Felix Nitsch Journal: tba. Volume: tba. DOI: 10.2139/ssrn.5320926 Status: Preprint available Licence: Open Access, CC BY 4.0 Executive Summary: Growing electricity price volatility has attracted significant investor interest in FOs, evidenced by over 226 GW of battery storage connection requests submitted to German transmission system operators in 2024 alone. Conversely, arbitrage by FOs tends to reduce price spreads as high prices are reduced by discharging and low prices are increased by charging. Thus, substantial uncertainty persists regarding how large-scale storage deployment will affect DAMs and individual economic viability under competitive market conditions. I address this problem by calibrating the AMIRIS model to a 2030 energy scenario for Germany, as defined in the Ariadne report2. My key methodological contribution extends AMIRIS with ML-based price forecasting capabilities that enable realistic simulation of competing FOs, thereby addressing a significant limitation of previous modelling approaches. With this ABM framework, I simultaneously capture individual FO economics and system-wide market dynamics through endogenous price formation. My analysis examines how forecasting accuracy influences operational decisions and competitive profitability. I find that imperfect ML forecasts can substantially reduce FO revenues, while operational strategy choices create significant performance differentials. Under accurate forecasting conditions, “risk-taking” strategies generate higher revenues than “risk-averse” approaches, although at the cost of increased cycling frequency. System-level analysis reveals a profitability plateau for homogeneous storage deployments between 4 to 8 GW installed capacity and 32 GWh total energy capacity. With favourable installation costs, annual returns can reach approximately 20% through DAM arbitrage alone. However, while heterogeneous storage technologies initially preserve their rev2The Ariadne report is widely recognised for its comprehensive analysis of potential pathways for the German energy transition towards climate neutrality by 2045 (Luderer, Kost, and Sörgel 2021). 3 Publications 73 F. Nitsch enue shares, individual profitability declines significantly once critical market penetration thresholds are exceeded. My findings indicate that current battery connection requests in Germany may exceed economically viable deployment levels in current market frameworks. The open-source modelling toolchain supports future research extensions which may focus on additional revenue streams such as ancillary services and cross-market optimisation strategies. Author Contributions: I am the lead author of this paper. The conceptual work was predominantly carried out by me, with input from my co-authors, Christoph Schimeczek and Valentin Bertsch. I was also responsible for software development, including implementing the AMIRIS extension package AMIRIS-PriceForecast. Together with Christoph Schimeczek, I developed the agent class PriceForecasterApi, which interfaces with the external forecasting model AMIRIS-PriceForecast via an application programming interface (API). I handled data curation, model execution, results validation, numerical analysis, and sensitivity analysis. All result visualisations were created by me and refined through discussions with my co-authors. Christoph Schimeczek and I were responsible for acquiring funding. I wrote the original draft, while both co-authors reviewed and edited it. 3 Publications 74 Nitsch et al. Manuscript 7/30 Figure 1: Modelling setup enabling agent-based electricity market analysis of competing flexibility options. 2.1 Electricity market modelling using AMIRIS AMIRIS is the Agent-based Market model for the Investigation of Renewable and Integrated energy Systems [38]. The electricity market model has been developed since 2008 and was published open source1 in late 2021 [41]. It is a powerful simulation based on the flexible framework FAME [42, 43]. The heart of AMIRIS is the simulation of the DAM revealing market dynamics and agent interactions [14] while considering different policy frameworks [44] and effects due to coupling of neighboring market zones [45]. AMIRIS has been back-tested for the German [46] and Austrian [47] DAM, which resulted in a good fit of simulated and historical electricity prices. Figure 12 in the Appendix provides an overview of the agents represented in AMIRIS (i.e. power plant operators, traders, flexibility operators, market operators and regulators) and their interactions via flows of information, energy, and money. Similar to other energy system models, users define and provide relevant input data [48]. In the context of AMIRIS this translates to power plant park structure, RE generation timeseries, demand data, and operational cost data. Central modelling outputs of AMIRIS are DAM electricity prices, as well as costs and revenues of market participants. FOs can choose between a “risk-taking” or more “risk-averse” bidding strategy when performing arbitrage at the DAM[15]. The consideration of further revenue streams for FOs, such as intraday markets, is not yet available. 1 gitlab.com/dlr-ve/esy/amiris (accessed on 20th June 2025) This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 81 Nitsch et al. Manuscript 8/30 Realistic analysis of FO competition requires modeling how operators make dispatch decisions under uncertainty, which fundamentally differs from optimization approaches that assume perfect market information. In DAM, FO profitability depends not only on actual price dynamics but on operators’ ability to anticipate and respond to these patterns. When multiple FOs compete, superior forecasting capabilities can lead to competitive advantages by better timing charge/discharge cycles. In contrast to energy optimization models where a central planner optimizes all units simultaneously, ABMs like AMIRIS require individual agents to make decisions based on imperfect information. This creates a critical need for electricity price forecasts provided to FO operators. This was previously addressed through simplified “naïve” forecasts eligible only for a single FO [49]. To address Gap 3 and thus enable FO competition (Gap 1), we substantially extended AMIRIS with a sophisticated forecasting system. Our key methodological contribution is the development of AMIRIS-PriceForecast [50], a dedicated machine-learning based forecasting module which runs in co-simulation with AMIRIS. As these algorithms have proven to deliver robust forecasts, even in future energy transition scenarios [51], we apply ML to derive accurate time series forecasts during AMIRIS runtime. This represents a significant advancement over both central-planner optimization models that assume perfect foresight and previous ABM studies that used oversimplified forecasting assumptions. The forecasting workflow, see also Figure 2, operates as follows: FO agents request price forecasts from a centralized forecasting agent (PriceForecasterApi) delivering the same forecast to each client. To reduce the number of calls to the external AMIRIS-PriceForecast, PriceForecasterApi retrieves cached forecasts meeting predefined accuracy criteria. When the accuracy of previous forecasts is insufficient or forecasts for uncovered time periods are needed, AMIRIS-PriceForecast loads a pre-trained model and generates time series predictions based on the specific forecast request. The coupling between the two components AMIRIS (Java based) and AMIRIS-PriceForecast (Python based) is accomplished using a standardized interface based on FastAPI2. The modular architecture and open-source implementation also ensures that researchers can extend the available forecasting algorithms to keep pace with advances in time series forecasting. 2 github.com/fastapi/fastapi (accessed on 20th June 2025) This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 82 Nitsch et al. Manuscript 9/30 Figure 2: Workflow of electricity price forecasting in AMIRIS and AMIRIS-PriceForecast. 2.2 Energy transition scenario Our study focuses on the German DAM zone. The data is based on the Ariadne scenario report [39], which is widely recognized for its comprehensive analysis of potential pathways for the German energy transition toward climate neutrality by 2045. The Ariadne project3, funded by the German Federal Ministry of Education and Research, employs multiple integrated modeling approaches to assess the feasibility, costs, and sectoral implications of different transformation pathways, making it a particularly robust foundation for energy system analysis. The economic and demographic assumptions underlying the Ariadne scenario report are based on the so-called “Middle of the Road” scenario among the Shared Socioeconomic Pathways as defined in [52]. The scenarios contain detailed assumptions about RE expansion, sector coupling mechanisms, hydrogen deployment strategies, and the systematic phase-out of fossil fuels across all economic 3 ariadneprojekt.de (accessed on 24th May 2025) This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 83 Nitsch et al. Manuscript 10/30 sectors. Specifically, we utilize the “mix” subscenario, which assumes the use of a mixed energy carrier portfolio (electricity, hydrogen, and synthetic green fuels) in final energy consumption, see also Table 1. This subscenario reflects a diversified approach to decarbonization where different sectors employ the most suitable clean energy carriers based on technical and economic considerations. We selected the year 2030 as our temporal focus, as many of the storage connection requests under consideration aim to be realized within this timeframe. Flexibilization of demand is already inherently integrated in the load profiles, reflecting the scenario’s consideration of demand-side management and behavioral adaptations. Thus, we add additional flexibility to the system which should represent the recent connection requests received by the German transmission system operators, thereby capturing the dynamic evolution of grid infrastructure requirements. Figure 3 provides an overview of the scenario projections on electricity supply and demand, illustrating the fundamental shifts in generation mix and consumption patterns anticipated for 2030. Table 1: Scenario parameters derived from Luderer et al. [39]. Parameter Value Unit Nuclear 0 GW Lignite 0 GW Hard coal 0 GW Natural gas 30.0 GW Hydrogen 15.3 GW Biomass 15.7 GW Run-of-river 12.6 GW PV 218.4 GW Wind onshore 127.2 GW Wind offshore 25.0 GW Capacities Other non-renewable 0.9 GW CO2 certificate costs 200.0 EUR/t Load 615.8 TWh/a Greenhouse gas reduction compared to 1990 65 % This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 84 Nitsch et al. Manuscript 11/30 Figure 3: Electricity generation and consumption for Germany according to the Ariadne scenario; adapted with permission from Luderer et al. [39]. 3 Results The results are organized in two main sections. Section 3.1 presents the methodological advancements developed to close gap 3, demonstrating the performance of our enhanced analytical framework. Section 3.2 examines the profitability of FOs in the future energy system scenario, providing insights to gap 1 and 2. 3.1 Methodological advances To analyze FOs in future electricity markets, we first evaluate the methodological improvements to AMIRIS. For this, we assess the accuracy of electricity price forecasts generated by the AMIRIS extension AMIRIS-PriceForecast. Typically, the FOs require at least 24 hours of electricity price forecasts to optimize their bidding schedule. Our approach employs a Temporal Fusion Transformer algorithm [53] for electricity price prediction. This architecture has demonstrated strong performance in forecasting electricity prices within future energy system scenarios [51]. The forecasting model integrated into AMIRIS-PriceForecast is iteratively called during AMIRIS runtime, generating DAM electricity price forecasts 𝑝 𝑡0 to 𝑝 𝑡23 based on previous electricity prices 𝑝 𝑡 ― 24 to 𝑝 𝑡 ― 1 and the residual load 𝑟𝑙 𝑡 ― 24 to 𝑟𝑙 𝑡23 . Figure 4 illustrates cumulative storage revenues achieved using different forecasting approaches, revealing the direct impact of forecast accuracy on FO performance. Perfect foresight represents the theoretical maximum revenue potential, while the baseline, a “naïve” TimeShift method using This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 85 Nitsch et al. Manuscript 12/30 the previous 24 hours as a forecast [54], captures less than 40% of this potential. Our analysis of various hyperparameter configurations demonstrates that model architecture significantly influences revenue outcomes. The optimal ML based prediction incorporates future covariates, achieving nearly 80% of potential revenues. This finding aligns with recent research showing substantial error reduction when integrating future covariates in time series forecasting [55]. While this enhanced performance requires significantly higher computational training4 costs, four to five times greater than alternative models, the training effort can be considered uncritical since the trained model is applicable to a wide range of simulation scenarios [51]. Each prediction call accounts for approximately 0.1 seconds including negligible overhead by the model coupling via FastAPI. Figure 4: Impact of forecasting model on cumulative storage revenue and required training effort. To examine how forecasting quality affects FO operation, we analyze a 1 MW, 5 MWh price-taking BSS with a round-trip efficiency (RTE) of 80% under two forecasting modes: perfect foresight, representing ex-post optimal operation, and model-endogenous ML based price forecasts. Each forecasting mode is combined with two operational strategies: risk-taking, which exploits all profitable price spreads, and risk-averse, which acts only on significant expected spreads. 4 Model training was performed on a system equipped with an Intel® Core™ i7-11850H processor (2.50 GHz) and 32 GB of RAM. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 86 Nitsch et al. Manuscript 13/30 Figure 5 illustrates the operational differences during a representative week. Perfect foresight enables optimal operation under the risk-taking strategy (Figure 5a), capturing all profitable arbitrage opportunities. The risk-averse strategy with perfect foresight (Figure 5b) results in reduced activity as some smaller price spreads remain unexploited, leading to missed charging and discharging opportunities (compare situations 5aI/5bI, or 5aII/5bII). ML based forecasting introduces prediction errors that create suboptimal storage patterns. Following the risk-taking strategy (Figure 5c), forecasting errors generate additional, often unprofitable charging and discharging events (compare situations 5aI/5cI). The risk-averse strategy with ML based forecasting (Figure 5d) leads to extended idle periods where profitable opportunities are missed due to forecast uncertainty (compare situations 5bI/5dI). As with perfect foresight, the risk-averse operation results in less activity compared to the risk-taking approach (compare situations 5cII/5dII). Figure 5: State of charge (blue area) and electricity price (black line) during a single week revealing the impact of storage strategy and electricity price forecast type. While Figure 5 provides a detailed view of a single week, Figure 6 reveals systematic patterns across all combinations by presenting an annual perspective in hourly resolution. Storage charging (red areas) occurs primarily during nighttime and midday low-price periods, while discharging (blue areas) results from peak prices during morning and evening. ML forecasting errors create noticeably noisier activity patterns, which are particularly evident in the risk-taking strategy (Figure 6b). The revenue analysis demonstrates substantial impacts from both strategy This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 87 Nitsch et al. Manuscript 14/30 choice and forecasting accuracy. Using perfect foresight as the baseline (risk-taking strategy, Figure 6a), the risk-averse strategy achieves approximately 79% of maximum revenues (Figure 6b). ML based forecasting reduces performance to 73% for risk-taking (Figure 6c) and 62% for risk-averse strategies (Figure 6d), see also Figure 13 in the Appendix. Cycling behavior also varies significantly across scenarios. Perfect foresight with risk-taking strategy results in 359 full charge cycles5 over the simulation year (nearly one per day). In contrast, the risk-averse approach reduces this number to 227 cycles. The risk-taking strategy joined with ML forecasting produces 381 cycles (indicating increased cycling due to forecast errors). This increased number of cycles could further impact profitability if cycling costs were considered. 207 cycles result when the riskaverse strategy is combined with ML forecasts, strengthening the earlier finding that forecast errors can lead to missed opportunities depending on the operational strategy employed. Figure 6: Storage activity, blue areas indicate charging, whereas red areas indicate discharging, during a full model year revealing impact by storage strategy and electricity price forecast type. 3.2 Profitability of flexibility options This section aims to identify optimal technical parameters for FOs, thus addressing Gap 2, and then examining how competitive dynamics affect profitability, which addresses Gap 1. To compare the profitability of differently scaled FOs, particularly BSS, we compute the return on investment (ROI), similar to the definition in [56]. ROI, see Equation (1), is calculated as the ratio of net 5 One full charge cycle indicates a complete charge and discharge of the storage, whether it occurs all at once or is accumulated over multiple partial cycles. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 88 Nitsch et al. Manuscript 15/30 cashflow 𝐶𝐹 𝑡 for year 𝑡 from earnings 𝑒 𝑡 minus costs 𝑐 𝑡 from (dis-)charging, see Equation (3), to the total installation costs 𝐶𝑜𝑠𝑡 0 , see Equation (4). The installation costs include the power-based converter costs 𝑐 𝑐𝑜𝑛𝑣𝑒𝑟𝑡𝑒𝑟 and the energy-based storage cost 𝑐 𝑠𝑡𝑜𝑟𝑎𝑔𝑒 . The result is expressed as a percentage where high values express superior profitability. This metric provides a quick paybackoriented perspective for initial investment screening. As ROI does not cover annualized capital costs, we solve Equation (2) for the Internal Rate of Return (IRR), as also used in [57] assuming an estimated operational lifetime and constant cashflow to provide an additional financial performance metric. 𝑅𝑂𝐼 = 𝐶𝐹 𝑡 = 1 𝐶𝑜𝑠𝑡 0 × 100 (1) 𝑁𝑃𝑉 = 𝐶𝐹 𝑡 = 1 × 𝑛 𝑡 = 1 1 (1 + 𝐼𝑅𝑅 ) 𝑡 ― 𝐶𝑜𝑠𝑡 0 = 0 (2) 𝐶𝐹 𝑡 = 𝑒 𝑡 ― 𝑐 𝑡 (3) 𝐶𝑜𝑠𝑡 0 = 𝑝𝑜𝑤𝑒𝑟 × 𝑐 𝑐𝑜𝑛𝑣𝑒𝑟𝑡𝑒𝑟 + 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 × 𝑐 𝑠𝑡𝑜𝑟𝑎𝑔𝑒 (4) Figure 7 presents ROI for homogeneous large-scale BSS in the 2030 scenario with fixed power at 8 GW and varying storage capacity (8 GWh to 64 GWh) under different cost assumptions. Figure 8 shows ROI when the capacity is fixed at 32 GWh and the storage power is varied from 1 GW to 12 GW. In all cases, converter costs range from 50 EUR/kW to 200 EUR/kW [58] while the storage costs range from 100 EUR/kWh to 400 EUR/kWh [59], and RTE is at 80%. The analysis reveals that annual ROI can exceed 18% under favorable cost conditions, with optimal performance occurring at 32 GWh total capacity and 4 to 8 GW installed power. Systems with low energy-to-power (E2P) ratios, e.g., 8 GW/8 GWh, or very high E2P ratios, e.g., 1 GW/32 GWh, show significantly lower ROI, indicating the importance of technical specifications adjusted to the energy system scenario. This can be explained by novel electricity price dynamics. Figure 15 in the Appendix provides a comprehensive scan of different storage systems and their resulting activity and price impact. We note that systems with low E2P ratios can charge and discharge rapidly, but cannot benefit from longer-duration price changes whereas systems with larger E2P generally profit from these periods. In these cases, however, storage costs represent the major share of total installation cost, thus causing maximum profitability at an intermediate E2P configurations. In contrast, specific system combinations, i.e., high-capacity and/or high-power This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 89 Nitsch et al. Manuscript 16/30 systems, impact market prices by elevating low prices during charging and suppressing high prices during discharging, thereby cannibalizing their collective revenue potential. A crucial distinction of our methodology is the endogenous modelling of electricity market prices, explicitly capturing how individual FO behavior influences market dynamics. This contrasts with studies such as [60], that treat prices as exogenous inputs, enabling us to identify saturation points where collective FO impacts reduce overall profitability. The impact of the ML training volume on ROI calculation is described in Figure 14 in the Appendix6. Figure 7: Return on investment (ROI) for homogeneous battery storage systems with fixed power (8 GW) and varying capacity. Figure 8: Return on investment (ROI) for homogeneous battery storage systems with fixed capacity (32 GWh) and varying power. To complement our profitability analysis based on ROI, we calculate the IRR for one of the best performing technical configuration, 6 GW/32 GWh. The IRR accounts for the time value of money 6 Figure 14 in the Appendix shows the effect of training length on ROI for BSS. When increasing training epochs from 1 to 50, we observe that the applied Temporal Fusion Transformer does not significantly improve if trained for more than 10 epochs. Compared to the analysis in Figure 4, we have increased the training data available to the model by a factor of five, thus resulting in 30 training scenarios with 8760 time steps each. Together with the used hyperparameter settings6, this allows the model to converge quickly. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 90 Nitsch et al. Manuscript 23/30 Figure 12: AMIRIS electricity market model. Figure 13: Storage revenue change based on applied strategy and forecast type. Figure 14: Impact of training length on return of investment for a 6 GW, 32 GWh battery storage system. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 97 Nitsch et al. Manuscript 24/30 Figure 15: State of charge (blue area) and electricity price (black line) during a single week revealing behavior and impact of different storage systems. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 98 Nitsch et al. Manuscript 25/30 References [1] C. Zöphel, S. Schreiber, T. Müller, and D. Möst, "Which Flexibility Options Facilitate the Integration of Intermittent Renewable Energy Sources in Electricity Systems?," Curr Sustainable Renewable Energy Rep, vol. 5, no. 1, pp. 37–44, 2018, doi: 10.1007/s40518018-0092-x. [2] S. Nyamathulla and C. Dhanamjayulu, "A review of battery energy storage systems and advanced battery management system for different applications: Challenges and recommendations," Journal of Energy Storage, vol. 86, p. 111179, 2024, doi: 10.1016/j.est.2024.111179. [3] S. Lieskoski, O. Koskinen, J. Tuuf, and M. Björklund-Sänkiaho, "A review of the current status of energy storage in Finland and future development prospects," Journal of Energy Storage, vol. 93, p. 112327, 2024, doi: 10.1016/j.est.2024.112327. [4] S. Enkhardt, Übertragungsnetzbetreibern liegen zum Jahreswechsel 650 Anschlussanfragen für große Batteriespeicher mit 226 Gigawatt vor. [Online]. Available: https://www.pv-magazine.de/2025/01/13/uebertragungsnetzbetreibern-liegen-zumjahreswechsel-650-anschlussanfragen-fuer-grosse-batteriespeicher-mit-226-gigawatt-vor/ [5] N. E. Koltsaklis and J. Knápek, "Assessing flexibility options in electricity market clearing," Renewable and Sustainable Energy Reviews, vol. 173, p. 113084, 2023, doi: 10.1016/j.rser.2022.113084. [6] M. E. Ölmez, I. Ari, and G. Tuzkaya, "A comprehensive review of the impacts of energy storage on power markets," Journal of Energy Storage, vol. 91, p. 111935, 2024, doi: 10.1016/j.est.2024.111935. [7] S. Quoilin, K. Kavvadias, A. Mercier, I. Pappone, and A. Zucker, "Quantifying selfconsumption linked to solar home battery systems: Statistical analysis and economic assessment," Applied Energy, vol. 182, pp. 58–67, 2016, doi: 10.1016/j.apenergy.2016.08.077. [8] C. Zhao, P. B. Andersen, C. Træholt, and S. Hashemi, "Grid-connected battery energy storage system: a review on application and integration," Renewable and Sustainable Energy Reviews, vol. 182, p. 113400, 2023, doi: 10.1016/j.rser.2023.113400. [9] Y. Zhao et al., "Energy storage for black start services: A review," Int J Miner Metall Mater, vol. 29, no. 4, pp. 691–704, 2022, doi: 10.1007/s12613-022-2445-0. [10] C. Straub, J. Maeght, C. Pache, P. Panciatici, and R. Rajagopal, "Congestion management within a multi-service scheduling coordination scheme for large battery storage systems," in 2019 IEEE Milan PowerTech, Milan, Italy, 2019, pp. 1–6. [11] J. Viehmann, "State of the German Short-Term Power Market," Z Energiewirtsch, vol. 41, no. 2, pp. 87–103, 2017, doi: 10.1007/s12398-017-0196-9. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 99 Nitsch et al. Manuscript 26/30 [12] O. Schmidt, S. Melchior, A. Hawkes, and I. Staffell, "Projecting the Future Levelized Cost of Electricity Storage Technologies," Joule, vol. 3, no. 1, pp. 81–100, 2019, doi: 10.1016/j.joule.2018.12.008. [13] R. Haas, C. Kemfert, H. Auer, A. Ajanovic, M. Sayer, and A. Hiesl, "On the economics of storage for electricity: Current state and future market design prospects," Wiley Interdisciplinary Reviews: Energy and Environment, vol. 11, no. 3, 2022, doi: 10.1002/wene.431. [14] F. Nitsch, M. Deissenroth-Uhrig, C. Schimeczek, and V. Bertsch, "Economic evaluation of battery storage systems bidding on day-ahead and automatic frequency restoration reserves markets," Applied Energy, vol. 298, p. 117267, 2021, doi: 10.1016/j.apenergy.2021.117267. [15] F. Nitsch, M. Wetzel, H. C. Gils, and K. Nienhaus, "The future role of Carnot batteries in Central Europe: Combining energy system and market perspective," Journal of Energy Storage, vol. 85, p. 110959, 2024, doi: 10.1016/j.est.2024.110959. [16] M. B. C. Salles, T. N. Gadotti, M. J. Aziz, and W. W. Hogan, "Potential revenue and breakeven of energy storage systems in PJM energy markets," Environmental science and pollution research international, vol. 28, no. 10, pp. 12357–12368, 2021, doi: 10.1007/s11356-018-3395-y. [17] M. Haugen et al., "Power market models for the clean energy transition: State of the art and future research needs," Applied Energy, vol. 357, p. 122495, 2024, doi: 10.1016/j.apenergy.2023.122495. [18] J. Geske and R. Green, "Optimal Storage, Investment and Management under Uncertainty: It is Costly to Avoid Outages!," The Energy Journal, vol. 41, no. 2, pp. 1–28, 2020, doi: 10.5547/01956574.41.2.jges. [19] J. Figgener et al., "The development of battery storage systems in Germany: A market review (status 2023)," 2023. [Online]. Available: http://arxiv.org/pdf/2203.06762v3 [20] A. D. Lamont, "Assessing the long-term system value of intermittent electric generation technologies," Energy Economics, vol. 30, no. 3, pp. 1208–1231, 2008, doi: 10.1016/j.eneco.2007.02.007. [21] Lion Hirth, "The market value of variable renewables: The effect of solar wind power variability on their relative price," Energy Economics, vol. 38, no. 0, pp. 218–236, 2013, doi: 10.1016/j.eneco.2013.02.004. [22] L. Reichenberg, T. Ekholm, and T. Boomsma, "Revenue and risk of variable renewable electricity investment: The cannibalization effect under high market penetration," Energy, vol. 284, p. 128419, 2023, doi: 10.1016/j.energy.2023.128419. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 100 Nitsch et al. Manuscript 27/30 [23] J. López Prol, K. W. Steininger, and D. Zilberman, "The cannibalization effect of wind and solar in the California wholesale electricity market," Energy Economics, vol. 85, p. 104552, 2020, doi: 10.1016/j.eneco.2019.104552. [24] J. López Prol and W.-P. Schill, "The Economics of Variable Renewable Energy and Electricity Storage," Annu. Rev. Resour. Econ., vol. 13, no. 1, pp. 443–467, 2021, doi: 10.1146/annurev-resource-101620-081246. [25] M. Kühnbach, J. Stute, and A.-L. Klingler, "Impacts of avalanche effects of price-optimized electric vehicle charging - Does demand response make it worse?," Energy Strategy Reviews, vol. 34, p. 100608, 2021, doi: 10.1016/j.esr.2020.100608. [26] E. Sperber, C. Schimeczek, U. Frey, K. K. Cao, and V. Bertsch, "Aligning heat pump operation with market signals: A win-win scenario for the electricity market and its actors?," Energy Reports, vol. 13, pp. 491–513, 2025, doi: 10.1016/j.egyr.2024.12.028. [27] A. Ensslen, P. Ringler, L. Dörr, P. Jochem, F. Zimmermann, and W. Fichtner, "Incentivizing smart charging: Modeling charging tariffs for electric vehicles in German and French electricity markets," Energy Research & Social Science, vol. 42, pp. 112–126, 2018, doi: 10.1016/j.erss.2018.02.013. [28] Sebastian Gottwalt, Wolfgang Ketter, Carsten Block, John Collins, and Christof Weinhardt, "Demand side management - A simulation of household behavior under variable prices," Energy Policy, vol. 39, no. 12, pp. 8163–8174, 2011, doi: 10.1016/j.enpol.2011.10.016. [29] J. Bistline et al., "Energy storage in long-term system models: a review of considerations, best practices, and research needs," Prog. Energy, vol. 2, no. 3, p. 32001, 2020, doi: 10.1088/2516-1083/ab9894. [30] R. Dumitrescu, R. Silvente, and P. Tankov, "Price impact and long-term profitability of energy storage," Oct. 2024. [Online]. Available: http://arxiv.org/pdf/2410.12495v1 [31] V. G. Lakiotis, C. K. Simoglou, and A. G. Bakirtzis, "A methodological approach for assessing the value of energy storage in the power system operation by mid-term simulation," Journal of Energy Storage, vol. 49, p. 104066, 2022, doi: 10.1016/j.est.2022.104066. [32] F. Scheller, R. Burkhardt, R. Schwarzeit, R. McKenna, and T. Bruckner, "Competition between simultaneous demand-side flexibility options: the case of community electricity storage systems," Applied Energy, vol. 269, p. 114969, 2020, doi: 10.1016/j.apenergy.2020.114969. [33] L. Deman, A. S. Siddiqui, C. Clastres, and Q. Boucher, "Day-ahead and Reserve Prices in a Renewable-based Power System: Adapting Electricity-market Design for Energy Storage," The Energy Journal, vol. 46, no. 2, pp. 67–98, 2025, doi: 10.1177/01956574241309557. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 101 Nitsch et al. Manuscript 28/30 [34] J. Sousa, J. Lagarto, and M. Fonseca, "The role of storage and flexibility in the energy transition: Substitution effect of resources with application to the Portuguese electricity system," Renewable Energy, vol. 228, p. 120694, 2024, doi: 10.1016/j.renene.2024.120694. [35] V. Sihvonen et al., "Combined utilization of electricity and thermal storages in a highly renewable energy system within an island society," Journal of Energy Storage, vol. 89, p. 111864, 2024, doi: 10.1016/j.est.2024.111864. [36] D. Zhao, M. Jafari, A. Botterud, and A. Sakti, "Strategic energy storage investments: A case study of the CAISO electricity market," Applied Energy, vol. 325, p. 119909, 2022, doi: 10.1016/j.apenergy.2022.119909. [37] N. Harder, A. Weidlich, and P. Staudt, "Finding individual strategies for storage units in electricity market models using deep reinforcement learning," Energy Inform, vol. 6, S1, 2023, doi: 10.1186/s42162-023-00293-0. [38] C. Schimeczek et al., "AMIRIS: Agent-based Market model for the Investigation of Renewable and Integrated energy Systems," JOSS, vol. 8, no. 84, p. 5041, 2023, doi: 10.21105/joss.05041. [39] G. Luderer, C. Kost, and D. Sörgel, "Deutschland auf dem Weg zur Klimaneutralität 2045 - Szenarien und Pfade im Modellvergleich," 2021. [40] F. Nitsch, U. Frey, and C. Schimeczek, AMIRIS-Scengen - A Scenario Generator for the Open Electricity Market Model AMIRIS: Zenodo, 2024. [41] K. Nienhaus et al., "AMIRIS. Agent-based Market model for the Investigation of Renewable and Integrated energy Systems.: https://gitlab.com/dlr-ve/esy/amiris/amiris," GitLab, 2021. [42] C. Schimeczek et al., "FAME-Core: An open Framework for distributed Agent-based Modelling of Energy systems," JOSS, vol. 8, no. 84, p. 5087, 2023, doi: 10.21105/joss.05087. [43] F. Nitsch, C. Schimeczek, U. Frey, and B. Fuchs, "FAME-Io: Configuration tools for complex agent-based simulations," JOSS, vol. 8, no. 84, p. 4958, 2023, doi: 10.21105/joss.04958. [44] U. J. Frey, M. Klein, K. Nienhaus, and C. Schimeczek, "Self-Reinforcing Electricity Price Dynamics under the Variable Market Premium Scheme," Energies, vol. 13, no. 20, 2020, doi: 10.3390/en13205350. [45] F. Nitsch and A. A. El Ghazi, "Energy systems analysis considering cross-border electricity trading: Coupling day-ahead markets in an agent-based electricity market model," 2023, doi: 10.5281/zenodo.10544676. [46] F. Maurer, F. Nitsch, J. Kochems, C. Schimeczek, V. Sander, and S. Lehnhoff, "Know Your Tools - A Comparison of Two Open Agent-Based Energy Market Models," in 2024 20th This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 102 Nitsch et al. Manuscript 29/30 International Conference on the European Energy Market (EEM), Istanbul, Turkiye, 2024, pp. 1–8. [47] F. Nitsch, C. Schimeczek, and S. Wehrle, "Back-testing the agent-based model AMIRIS for the Austrian day-ahead electricity market," 2021. [Online]. Available: https://zenodo.org/ record/5726738 [48] I. M. Müller, "Feature selection for energy system modeling: Identification of relevant time series information," Energy and AI, vol. 4, p. 100057, 2021. [49] F. Nitsch and C. Schimeczek, "Comparison of electricity price forecasting methods for use in agent-based energy system models," Vienna, 2023. [Online]. Available: https://elib.dlr.de/ 194021/ [50] F. Nitsch and C. Schimeczek, "AMIRIS-PriceForecast," 2025, doi: 10.5281/zenodo.14907870. [51] F. Nitsch, C. Schimeczek, and V. Bertsch, "Applying machine learning to electricity price forecasting in simulated energy market scenarios," Energy Reports, vol. 12, pp. 5268–5279, 2024, doi: 10.1016/j.egyr.2024.11.013. [52] H. Lee et al., "IPCC, 2023: Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Core Writing Team, H. Lee and J. Romero (eds.)]. IPCC, Geneva, Switzerland," 2023. [53] B. Lim, S. Ö. Arık, N. Loeff, and T. Pfister, "Temporal fusion transformers for interpretable multi-horizon time series forecasting," International Journal of Forecasting, 2021. [54] R. J. Hyndman and G. Athanasopoulos, "Forecasting: principles and practice," Melbourne, 2018. [55] R. Theiler, L. von Krannichfeldt, G. Sansavini, M. F. Howland, and O. Fink, "Integrating the Expected Future in Load Forecasts with Contextually Enhanced Transformer Models," 2024. [56] M. Naumann, R. C. Karl, C. N. Truong, A. Jossen, and H. C. Hesse, "Lithium-ion Battery Cost Analysis in PV-household Application," Energy Procedia, vol. 73, pp. 37–47, 2015, doi: 10.1016/j.egypro.2015.07.555. [57] Y. Yang et al., "Life cycle economic viability analysis of battery storage in electricity market," Journal of Energy Storage, vol. 70, p. 107800, 2023, doi: 10.1016/j.est.2023.107800. [58] NREL, "Annual Technology Baseline: Utility-Scale Battery Storage," 2025. [Online]. Available: https://atb.nrel.gov/electricity/2024/utility-scale_battery_storage [59] W. Cole and A. Karmakar, "Cost Projections for Utility-Scale Battery Storage: 2023 Update," 2023. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 103 Nitsch et al. Manuscript 30/30 [60] Petr Spodniak, Valentin Bertsch, and Mel Devine, "The Profitability of Energy Storage in European Electricity Markets," The Energy Journal, vol. 42, no. 5, 2021, doi: 10.5547/01956574.42.5.pspo. [61] E. Kraft, M. Russo, D. Keles, and V. Bertsch, "Stochastic optimization of trading strategies in sequential electricity markets," European Journal of Operational Research, vol. 308, no. 1, pp. 400–421, 2023, doi: 10.1016/j.ejor.2022.10.040. [62] S. Karhinen and H. Huuki, "Private and social benefits of a pumped hydro energy storage with increasing amount of wind power," Energy Economics, vol. 81, pp. 942–959, 2019, doi: 10.1016/j.eneco.2019.05.024. [63] D. S. Mallapragada, N. A. Sepulveda, and J. D. Jenkins, "Long-run system value of battery energy storage in future grids with increasing wind and solar generation," Applied Energy, vol. 275, p. 115390, 2020, doi: 10.1016/j.apenergy.2020.115390. [64] D. Keles, J. Scelle, F. Paraschiv, and W. Fichtner, "Extended forecast methods for dayahead electricity spot prices applying artificial neural networks," Applied Energy, vol. 162, pp. 218–230, 2016, doi: 10.1016/j.apenergy.2015.09.087. [65] C. Fraunholz, E. Kraft, D. Keles, and W. Fichtner, "Advanced price forecasting in agentbased electricity market simulation," Applied Energy, vol. 290, p. 116688, 2021, doi: 10.1016/j.apenergy.2021.116688. [66] S. Pfenninger, A. Hawkes, and J. Keirstead, "Energy systems modeling for twenty-first century energy challenges," Renewable and Sustainable Energy Reviews, vol. 33, pp. 74– 86, 2014, doi: 10.1016/j.rser.2014.02.003. [67] W. Antweiler and F. Muesgens, "The new merit order: The viability of energy-only electricity markets with only intermittent renewable energy sources and grid-scale storage," Energy Economics, vol. 145, p. 108439, 2025, doi: 10.1016/j.eneco.2025.108439. [68] F. Roques and D. Finon, "Adapting electricity markets to decarbonisation and security of supply objectives: Toward a hybrid regime?," Energy Policy, vol. 105, pp. 584–596, 2017, doi: 10.1016/j.enpol.2017.02.035. [69] D. Keles, P. Jochem, R. McKenna, M. Ruppert, and W. Fichtner, "Meeting the Modeling Needs of Future Energy Systems," Energy Technology, 5(7), pp. 1007–1025, 2017, doi: 10.1002/ente.201600607. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5320926 Preprint not peer reviewed F. Nitsch 3 Publications 104 F. Nitsch 3.5 Synthesis The presented publications form a coherent body of research that systematically investigates the economic feasibility of FOs from multiple complementary perspectives. This research progresses from an initial exploration of individual technologies to comprehensive analyses that include system-wide competition effects. Each publication addresses the limitations identified in previous work, while pursuing the overarching research targets defined in Section 1.2. The publications are divided into “main papers” (Sections 3.1 to 3.4), “supplementary papers” (Appendix A), and “supplementary non-peer-reviewed papers” (Appendix B). Table 3.1 shows the individual contributions of each paper to the research targets. This section also places the presented publications in context and demonstrates their systematic contribution to a unified research narrative. Table 3.1: Contribution of publications to research targets. Main paper Supplementary Paper Peer-review None Target I II III IV A1 A2 A3 A4 A5 B1 B2 1.1 Strategies3x x x 1.2 Parameters4x x x x 1.3 Competition5x x 2.1 Features6(x) x x x x x x x 2.2 Open science7x x x x x x x x x 3Identify reliable operational strategies for FOs in future electricity market scenarios. 4Evaluate how technical specifications influence FO refinancing potential. 5Quantify the impact of FO competion on their profitability. 6Expand ABM to capture individual FO economics and impact on system dynamics. 7Develop modular open-source software packages. 3 Publications 105 F. Nitsch 3.5.1 Individual technology analysis In Paper I, I introduced a representation of the aFRR in AMIRIS, allowing a battery storage operator to generate multi-market revenues beyond DAM arbitrage. This work investigates the influence of storage parameters on the revenue potential for a future energy system. I found that revenues for FO operators across the two simulated markets are highest for short-term orientated systems. However, explicit feedback from FO operation on the overall energy system was not considered, a major limitation addressed in subsequent papers. Back-testing was performed for the German market zone in 2019 achieving good agreement with historical market data. This paper allowed for identifying the most promising technical specifications that provide the highest revenue, directly contributing to research target 1.2. It also advances target 2.1 by extending AMIRIS with a mechanism to include revenues from the aFRR market. Paper II focuses on high-temperature heat storage, namely Carnot batteries, rather than battery storage systems. To extend the analysis to endogenously modelled systemscale storage systems, I linked the ESOM REMix (Wetzel et al. 2024) with AMIRIS. The optimised power plant park from REMix is used as scenario parameterisation for AMIRIS to investigate the economic potential of a Carnot battery storage. For this model coupling, I developed the open workflow iog2x (Nitsch, Schimeczek, Wetzel, et al. 2023), a tool based on the open workflow manager ioproc (Fuchs et al. 2020). Like the first main paper, this work adds insights to research targets 2.1 and 2.2. However, this research focuses on energy system with much higher RE shares and concentrates on Carnot battery storage systems providing flexibility. By revealing how the operational strategy and technical specifications of Carnot battery storage impact revenue potential, this paper advances research targets 1.1 and 1.2 respectively. 3.5.2 Methodological innovation for competition analysis While these first two main papers (Section 3.5.1) provide valuable information on the operation of two dedicated FO technologies, battery storage and Carnot battery storage, they share important limitations. Both model only a single FO in the system, a smallscale single battery storage (Paper I), or a large-scale Carnot battery storage (Paper II). Additionally, I did not explicitly model competition among FOs, thus neglecting potential cannibalisation effects that can limit the profitability of such systems. Endogenous competition between agents is challenging to model in ESA, see elaboration to Figure 2.6, but provides a critical factor for better assessing the FO performance in future energy systems. Therefore, I worked to represent such competing behaviour on DAMs by enhancing the 3 Publications 106 F. Nitsch competitive dynamics modeled in this research and ultimately impact the profitability of individual FOs. Although the methodological approaches developed can be applied to multiple market zones, this work focuses primarily on the German energy system. The energy transition progresses at different rates in neighbouring market zones, which can create novel cross-border effects that are not captured in the analysis. Market coupling mechanisms provide some representation of these interactions, but the detailed modelling of heterogeneous transition pathways across interconnected European markets remains beyond the scope of this research. The analysis presents a limited picture of FO revenue potential due to its primary focus on DAM simulation. Except for Paper I, which incorporated a representation of an automatic Frequency Restoration Reserves (aFRR) market, the research does not employ a multi-market approach that would capture a wider spectrum of revenue opportunities available to FOs. Existing literature suggests that overall revenue increases when multimarket arbitrage across markets is pursued (Agrela, Rezende, and Soares 2022). However, consistent with the cannibalisation findings presented in Paper IV, these additional revenue streams would likely face similar profitability limits as FO market penetration increases and competition intensifies (Deman et al. 2025). It is also worth noting that the electricity market modeling in AMIRIS does not fully capture the possibilities of strategic bidding, such as portfolio effects or complex bids. With increasing shares of FOs, this could become an important aspect for future studies (Signer et al. 2025). Ongoing discussions about electricity market design adaptations could also substantially impact FO economics. Besides the ongoing harmonisation from hourly to 15-minute DAM clearing intervals (European Parliament and Council of the European Union 2019), discussions are about market zone splitting (Mörtenkötter et al. 2025) or transition to nodal pricing (Knörr, Bichler, and Dobos 2025), new RE remuneration schemes (Schlecht, C. Maurer, and Hirth 2024), dynamic grid tariffs (Stute and Klobasa 2024), or complete reworks of market design, such as capacity markets (Ölmez, Ari, and Tuzkaya 2024). Each of these modifications could fundamentally affect the revenue situation for FOs. Modellers should also be aware of the biases inherent in any of the presented ESA model, which is especially important when using the ML-based electricity price forecasts. 4.3 Conclusions The goal of this thesis was to contribute to a better understanding of FOs in energy systems with high shares of RE. Through systematic application of ABM to investigate electricity markets, this research provides significant contributions to the identified 4 Discussion and conclusions 113 F. Nitsch research targets while advancing both methodological capabilities and practical insights for the energy transition. Research target 1.1 Identify operational strategies for FOs that perform reliably in future electricity market scenarios with increasingly high RE shares The analysis of operational strategies reveals that FO performance in RE-dominated systems depends critically on both individual strategy selection and the broader system context with significant implications for system-wide deployment. Paper II establishes an understanding of operational strategy impacts by comparing profit maximisation versus system cost minimisation approaches for a single large-scale FO accounting for its own price impacts. As anticipated, profit maximisation generally outperforms system cost minimisation in terms of economic revenue potential when the storage system is the sole large-scale FO in the system. However, the research also reveals a dependency on the storage system’s role within the energy system. The analysis demonstrates that the storage system achieves higher revenue per installed MW when alternative ways of providing flexibility, such as grid extension, are unavailable. This finding can be interpreted as an early indication of cannibalisation effects, where the presence of competing flexibility sources reduces individual FO profitability. The strategic implications suggest that FO operational strategies must account not only for their own technical capabilities but also for the availability and deployment of alternative system flexibility resources. Paper IV advances the understanding of operational strategies by analysing their performance under competitive conditions with multiple FOs. The research demonstrates that storage strategy selection has substantial impacts on both overall revenue generation and operational characteristics such as charging cycles. Quantitatively, risk-averse strategies achieve approximately 20% lower profits compared to risk-taking strategies. The competitive analysis focuses primarily on risk-taking strategies to investigate cannibalisation effects. A fundamental challenge emerges when FOs collectively influence market prices yet possess only limited awareness of their own price impacts when optimising their individual strategy. This poses a problem, particularly for short-duration storage systems (low E2P ratios) that can lead to significant cumulative market effects. These short-duration systems would benefit substantially from more sophisticated operational strategies that explicitly account for their collective price impacts during dispatch decisions. The findings reveal that reliable FO operational strategies for high RE scenarios must incorporate several key elements, and that a trade-off must be struck between maximising revenues and minimising risk. This includes enhanced price impact awareness 4 Discussion and conclusions 114 F. Nitsch that account for their own market influence, and approaches that respond to changing competitive conditions as FO market penetration increases. Furthermore, the research suggests that operation of FOs cannot be considered independently of technical specifications and market structure. The interaction between strategy selection, technical parameters (particularly E2P ratios), and competitive environment creates challenges that traditional approaches accounting for only single or few FOs cannot adequately address. The evolution from individual optimisation (Paper II) to competitive analysis (Paper IV) demonstrates that operational strategies that appear optimal in an isolated analysis may prove suboptimal when deployed at scale. Research target 1.2 Evaluate how technical specifications (e.g., capacity & power) of FOs influence refinancing potential under varying cost assumptions The progression across papers demonstrates that optimal technical configurations depend critically on market context and competition levels. Paper I establishes foundational insights for a single FO system operating in a multi-market setting. The analysis reveals that small E2P ratios perform optimally for DAM arbitrage with aFRR market participation, as these configurations best exploit short-term price volatility across both markets. The research quantifies that round-trip-efficiency (RTE) improvements of 1% translate to approximately 2.5% additional revenue, highlighting the high impact of technical performance on economic outcomes. However, this analysis focuses solely on revenue generation without comprehensive cost assessment, limiting insights into actual refinancing potential. Paper II shifts focus to system-scale storage technologies, specifically Carnot batteries, revealing that storage costs represent the most critical factor for revenue generation. The analysis identifies that energy specific investment costs should be below 35 EUR/kWh for economic viability. This marks an ambitious target that highlights the challenge of making large-scale high-temperature storage economically competitive. RTE improvements emerge as a valid secondary optimisation target, consistent with Paper I findings. Importantly, the revenue situation proves sensitive to scenario assumptions, particularly regarding competing flexibility, such as grid expansion, which directly affects the Carnot battery’s market position and revenue potential. Paper IV provides the most comprehensive insights by analysing technical specifications within a competitive market scenario that also account for endogenous price impacts. This study reveals different results to those found in earlier studies. In contrast to Paper I’s conclusion favoring low E2P ratios, the competitive analysis demonstrates that higher E2P ratios tend to be more profitable. This reversal occurs because longer-duration storage can participate actively across more hours of operation, providing greater market 4 Discussion and conclusions 115 F. Nitsch flexibility. The competition analysis reveals that the number of profitable arbitrage hours remains similar across different scenario configurations, but higher E2P ratios enable storage systems to capture value across extended periods. However, the research identifies a profitability plateau as long-term storage systems are surpassed by medium-term configurations. This suggests that E2P ratios have to be aligned with the existing electricity price dynamics in order to effectively provide operational flexibility. Consistent with findings in Paper II, storage cost reductions generally provide greater profitability improvements than converter cost reductions, emphasising that energy storage capacity costs are a critical parameter to FO economic viability. The competitive analysis further reveals that either long system lifetimes of at least 10 years or significant cost reductions are necessary for profitability when FOs compete with other market participants. Collectively, these findings demonstrate that optimal FO technical specifications are not static parameters but depend critically on market structure, competition levels, and system-wide deployment scenarios. The evolution from favoring short-duration storage systems in individual analysis to medium-duration systems in competitive environments illustrates the importance of considering market context in technology development and investment decisions. The consistent emphasis on storage cost reduction across multiple studies provides clear guidance for research and development priorities in FO technologies. Research target 1.3 Quantify the impact of increasing market penetration of competing FOs on their profitability in DAM The evolution towards understanding competition effects represents a central achievement of this thesis. Paper I analysed variants of a single small-scale price-taking storage system, providing valuable insights into revenue potential while explicitly acknowledging the limitation of not accounting for competition among FOs. Paper II advanced this research by examining a single grid-scale storage system that accounts for its own price impacts, yet still did not capture interactions among multiple FOs, thus this critical gap remained. Recognising this limitation, Paper III advanced time series forecasting capabilities for ABM in energy transition scenarios. This work established the technical foundation necessary to investigate competition among FOs by enabling accurate price forecasts in RE-dominated markets with fundamentally different dynamics than historical years. The methodological innovations in this regard are then continued in Paper IV, where extensions to AMIRIS, particularly the design and implementation of AMIRISPriceForecast, enable simulation of competing FOs. This is achieved by endogenously modelling price feedbacks from multiple FOs which affect their revenue potential. This capability allows, for the first time, the analysis of both homogeneous and heterogeneous 4 Discussion and conclusions 116 F. Nitsch FOs technologies competing in future energy transition scenarios. The results reveal significant cannibalisation effects that have strong implications for FO deployment strategies. Once a certain market penetration threshold is reached, dependent on the specific power and capacity specifications of the FO technologies, both collective and individual revenues decline significantly. This finding directly challenges current trends and interest for large-scale FO deployments which often neglect their collective feedback on DAM prices. My results thus contribute critical insights for policy and investment decisions. Most significantly, my analysis suggests that connection requests for grid-scale storage of more than 200 GW submitted to German transmission system operators (Enkhardt 2025) are very likely far beyond what can be refinanced from DAM arbitrage alone. This conclusion has immediate practical relevance for energy system planning and highlights the importance of considering competition effects in FO deployment strategies. Research target 2.1 Expand ABM to simultaneously capture both individual FO economics and their collective impact on system dynamics This thesis achieves a fundamental advancement in ABM capabilities by systematically developing the methodological foundations necessary to model both individual FO economics and their collective system-level impacts. The progression across papers demonstrates a clear evolution from isolated individual analysis to comprehensive multi-agent competition modelling. Paper I introduced a aFRR market representation as an external optimisation model which uses AMIRIS simulation results. Agents bid with their opportunity costs across multiple markets, thereby facilitating multi-market revenue simulation. However, this initial implementation has some critical limitations, such as missing feedback from the external aFRR model to the DAM, and the focus on a single small-scale storage device. Paper II addressed the feedback limitation through integrated FO implementation that captures system-scale storage impacts on market dynamics. By enabling FOs to account for their own price impacts, the agent aims to maximise profits or minimise total system cost. This establishes a closed feedback loop that simultaneously analyses individual economics and system-level storage impacts which is a crucial advance towards realistic FO modelling. The automated model coupling linking energy systems optimisation model (ESOM) and ABM further enhances this contribution, addressing the complementary limitations where AMIRIS cannot define cost-optimal power plant parks while REMix cannot fully capture market dynamics. However, the challenge of modelling multiple competing FOs remained unresolved (see Section 2.2.1.2). Paper III provides the 4 Discussion and conclusions 117 F. Nitsch methodological breakthrough that enables endogenous competition simulation. FOs require sophisticated price forecasts for schedule optimisation, but traditional approaches fail when multiple FOs act on identical signals. The packages focapy and AMIRISScengen present powerful tools for this analysis. By quantifying errors of ML-based electricity price forecasts specifically in future energy system scenarios, this work proves that ML can provide enhanced and accurate electricity price forecasts. This implementation therefore represents a fundamental methodological cornerstone that finally prepares AMIRIS to model competing FOs. Paper IV applies all these methodological advances by implementing ML-based price forecasting for FO agents within AMIRIS. This integration achieves the primary objective of analysing FO competition modelling while capturing their impact on DAM. This enables realistic analysis of future energy transition scenarios with increasing FO market penetration. Further emphasis was placed in the general software publications of AMIRIS (see Paper A1) and FAME-Io (see Paper A2). Through this systematic progression, the thesis successfully transforms AMIRIS from a model capable of analysing only isolated individual FOs to a comprehensive platform that simultaneously captures individual economics and collective system dynamics. This advancement enables, for the first time, realistic ABM assessment of FO deployment accounting for both technical performance and market competition effects. Research target 2.2 Develop modular open-source software packages to enhance reproducibility and facilitate comparative assessment of FOs This thesis makes substantial open science contributions to the ESA modelling community through consistent development of modular software packages, enhancement of existing tools, and provision of data. The progression across the presented papers demonstrates increasing sophistication in software development practices and community engagement. In Paper I, a pre-FAME instance of AMIRIS was used. Although, this version is not directly compatible with the current versions of the model, it contributed important foundational elements, such as backtesting data for the German DAM in 2019, which is now part of the open AMIRIS-Examples (Nienhaus et al. 2025). Similarly, supplementary Paper B1 extended this data contribution with Austrian DAM backtesting data for 2019, enhancing the empirical validation resources available. Paper II marked a significant advancement in open-source contribution quality as it applies a fully open AMIRIS model instance. Additionally, the iog2x workflow provides a generic coupling mechanism that can convert any model producing GDX result files, such as REMix, to be accessible to AMIRIS. This tool extends beyond the specific use case in the paper to enable broader 4 Discussion and conclusions 118 F. Nitsch model coupling applications (Sarfarazi, Sasanpour, and Cao 2023; Torralba-Díaz et al. 2024; Kochems et al. 2024). Paper III holds two major software contributions through two comprehensive packages. AMIRIS-Scengen provides a flexible scenario generator for AMIRIS that serves multiple purposes. It allows multi-scenario analysis and also generates training data sets for ML applications. focapy delivers a complete software package for training and inference in ML time series prediction. It fills a critical gap in the ESA domain, as it is targeted for users applying and working with time series in ESA models. Paper IV features comprehensive AMIRIS model extensions. The AMIRIS-PriceForecast extension makes time series forecasts available during AMIRIS model runtime, therefore fundamentally improving the model’s forecasting capabilities. It is implemented with a flexible FastAPI-based interface that enables use both within and outside the AMIRIS ecosystem. The high performance of these tools enables the demanding simulation of competition among multiple FOs, opening new research possibilities for the community. Beyond the main papers, the supplementary contributions demonstrate sustained commitment to research software and community building. Given that modelling FOs is a critical challenge for many ABM (Schimeczek, Khanra, and Signer 2025), these contributions address pressing community needs. The model comparison with ASSUME provides valuable benchmarking insights (see Paper A3), while the market coupling algorithm enables endogenous simulation of multiple market zones in AMIRIS (see Paper A4). I also followed community engagement efforts (Nitsch, Schimeczek, Nienhaus, et al. 2025), including multiple conference presentations1, enhanced and peer-reviewed model documentation (see Paper A1), and establishment of a weekly open forum2. These initiatives have fostered a growing external community and attracted external code contributions, demonstrating the practical value and adoption of the developed tools. Furthermore, the FAME framework provides a powerful and flexible software suite for ABM in the ESA domain. My specific work in Paper A2 on FAME-Io follows high software development standards and findable, accessible, interoperable, reusable (FAIR) principles, with recent enhancements in Paper B2 including metadata capabilities that improve interoperability and scientific reproducibility. Collectively, these software contributions represent a comprehensive selection of tools that not only enable the specific research presented in this thesis but also provide the community with robust, well-documented, and interoperable software packages that will facilitate future research. 1https://zenodo.org/communities/amiris/ 2https://gitlab.com/dlr-ve/esy/amiris/amiris/-/wikis/Community/Support 4 Discussion and conclusions 119 F. Nitsch 4.4 Outlook The research presented in this thesis opens up several avenues for future investigation, both in terms of content and methodological focus. Future work should examine the impact of FOs at different grid levels, in particular the interaction between transmission system operators and distribution system operators. The contribution of FOs, e.g. large-scale battery storage systems, to reducing redispatch costs could also be another promising research direction. This would include the optimal allocation of battery resources in the electricity grid and their systemic impact, such as system costs and grid stability. Investigating the regulatory aspects that either enable or constrain the participation of batteries in system services could also be an important research direction. While energy-only markets have traditionally dominated European electricity systems, the increasing penetration of RE has intensified discussions on the need for explicit capacity remuneration mechanisms (Strbac et al. 2021). Changes in market design, in particular the introduction of a capacity market, demand a thorough investigation of their impact on FOs, especially on their investment and operation and how these changes are implemented across different market zones (Bucksteeg, Spiecker, and Weber 2019). The interactions between such mechanisms and the provision of flexibility require further in-depth analysis. Finally, expanding the range of scenarios would allow for the exploration of more diverse energy transition pathways, including different rates of electrification across sectors and different technology mixes. From a methodological perspective, several extensions would strengthen the research presented. The introduction of an intraday market would allow the investigation of additional revenue streams for FOs allowing for a more complete assessment of profitability. Advancing the competition modelling between FOs would further benefit the understanding of avalanche effects and how they can be mitigated. Extending the available forecasting methods to better represent uncertainty and increase the realism of agent decisions would improve model validity. This could include stochastic optimisation techniques that better capture the probabilistic nature of uncertainty in electricity price forecasts. Finally, extending both the spatial and temporal scope could contribute to a broader understanding of new situations due to climate change and assess the contribution of FOs to system stability. 4 Discussion and conclusions 120 References Agrela, João Carlos, Igor Rezende, and Tiago Soares (2022). „Analysis of battery energy storage systems participation in multi-services electricity markets“. In: 2022 18th International Conference on the European Energy Market (EEM), pp. 1–6. doi:10. 1109/EEM54602.2022.9921164. Alizadeh, Mohammed I, M Parsa Moghaddam, Nima Amjady, Pierluigi Siano, and Mohammed K Sheikh-El-Eslami (2016). „Flexibility in future power systems with high renewable penetration: A review“. In: Renewable and Sustainable Energy Reviews 57, pp. 1186–1193. doi:10.1016/j.rser.2015.12.200. Alon, Ilan, Min Qi, and Robert J. Sadowski (2001). „Forecasting aggregate retail sales: a comparison of artificial neural networks and traditional methods“. In: Journal of retailing and consumer services 8.3, pp. 147–156. doi:10.1016/S09696989(00) 00011-4. Alpaydin, Ethem (2021). Machine learning. MIT press. isbn: 9780262043793. Amor, Souhir Ben, Thomas Möbius, and Felix Müsgens (2024). Bridging an energy system model with an ensemble deep-learning approach for electricity price forecasting.doi: 10.48550/arXiv.2411.04880. Azevedo, Inês, Christopher Bataille, John Bistline, Leon Clarke, and Steven Davis (2021). „Net-zero emissions energy systems: What we know and do not know“. In: Energy and Climate Change 2, p. 100049. issn: 2666-2787. doi:10.1016/j.egycc.2021.100049. Babatunde, Olubayo Moses, Josiah L Munda, and YJER Hamam (2020). „Power system flexibility: A review“. In: Energy Reports 6, pp. 101–106. doi:10.1016/j.egyr.2019. 11.048. Bale, Catherine S.E., Liz Varga, and Timothy J. Foxon (2015). „Energy and complexity: New ways forward“. In: Applied Energy 138, pp. 150–159. issn: 0306-2619. doi:10. 1016/j.apenergy.2014.10.057. Baran, Ágnes, Sebastian Lerch, Mehrez El Ayari, and Sándor Baran (2020). „Machine learning for total cloud cover prediction“. In: Neural Computing and Applications 2020 33:7 33.7, pp. 2605–2620. issn: 1433-3058. doi:10.1007/S00521-020-05139-4. Barbosa, Aglaucibelly Maciel, Paulo Rotella Junior, Luiz Célio Souza Rocha, Anrafel de Souza Barbosa, and Ivan Bolis (2024). „Optimization methods of distributed hybrid power systems with battery storage system: a systematic review“. In: Journal of Energy Storage 97, p. 112909. doi:10.1016/j.est.2024.112909. 121 F. Nitsch Barker, Michelle et al. (2022). „Introducing the FAIR Principles for research software“. In: Scientific Data 9.1, p. 622. doi:10.1038/s41597-022-01710-x. Barnett, Harold J (1950). Energy uses and supplies, 1939, 1947, 1965. Tech. rep. Bureau of Mines, Washington, DC (USA). Bellman, Richard (1957). Dynamic Programming. Dover Publications. isbn: 9780486428093. Bernath, Christiane, Gerda Deac, and Frank Sensfuß (2021). „Impact of sector coupling on the market value of renewable energies – A model-based scenario analysis“. In: Applied Energy 281, p. 115985. issn: 0306-2619. doi:10.1016/j.apenergy.2020.115985. Bertram, Christoph et al. (2021). „Energy system developments and investments in the decisive decade for the Paris Agreement goals“. In: Environmental Research Letters 16.7, p. 074020. doi:10.1088/1748-9326/ac09ae. Bessa, Ricardo, Carlos Moreira, Bernardo Silva, and Manuel Matos (2019). „Handling renewable energy variability and uncertainty in power system operation“. In: Advances in Energy Systems: The Large-scale Renewable Energy Integration Challenge, pp. 1– 26. doi:10.1002/wene.76. Bjerregøard, Mathias Blicher, Jan Kloppenborg Møller, and Henrik Madsen (2021). „An introduction to multivariate probabilistic forecast evaluation“. In: Energy and AI, p. 100058. doi:10.1016/j.egyai.2021.100058. Bogdanov, Dmitrii, Ashish Gulagi, Mahdi Fasihi, and Christian Breyer (2021). „Full energy sector transition towards 100% renewable energy supply: Integrating power, heat, transport and industry sectors including desalination“. In: Applied Energy 283, p. 116273. doi:10.1016/j.apenergy.2020.116273. Bogdanov, Dmitrii, Manish Ram, et al. (2021). „Low-cost renewable electricity as the key driver of the global energy transition towards sustainability“. In: Energy 227, p. 120467. doi:10.1016/j.energy.2021.120467. Böse, Joos-Hendrik, Valentin Flunkert, Jan Gasthaus, Tim Januschowski, Dustin Lange, David Salinas, Sebastian Schelter, Matthias Seeger, and Yuyang Wang (2017). „Probabilistic demand forecasting at scale“. In: Proceedings of the VLDB Endowment 10.12, pp. 1694–1705. doi:10.14778/3137765.3137775. Brown, Tom et al. (n.d.). PyPSA: Python for Power System Analysis. Version 0.34.0. doi: 10.5334/jors.188. Bucksteeg, Michael, Stephan Spiecker, and Christoph Weber (2019). „Impact of Coordinated Capacity Mechanisms on the European Power Market“. In: The Energy Journal 40.2, pp. 221–264. doi:10.5547/01956574.40.2.mbuc. References 122 F. Nitsch Löschel, Andreas, Dirk Rübbelke, Wolfgang Ströbele, Wolfgang Pfaffenberger, and Michael Heuterkes (2020). Energiewirtschaft: Einführung in Theorie und Politik. Walter de Gruyter GmbH & Co KG. Luderer, Gunnar, Christoph Kost, and Dominika Sörgel (2021). Deutschland auf dem Weg zur Klimaneutralität 2045 - Szenarien und Pfade im Modellvergleich, (AriadneReport). Potsdam Institute for Climate Impact Research. doi:10.48485/pik.2021. 006. Lund, Peter D, Juuso Lindgren, Jani Mikkola, and Jyri Salpakari (2015). „Review of energy system flexibility measures to enable high levels of variable renewable electricity“. In: Renewable and Sustainable Energy Reviews 45, pp. 785–807. issn: 1364-0321. doi: 10.1016/j.rser.2015.01.057. Ma, Tieju and Yoshiteru Nakamori (2009). „Modeling technological change in energy systems – From optimization to agent-based modeling“. In: Energy 34.7, pp. 873–879. issn: 0360-5442. doi:10.1016/j.energy.2009.03.005. Madan, Rishabh and Partha Sarathi Mangipudi (2018). „Predicting computer network traffic: a time series forecasting approach using DWT, ARIMA and RNN“. In: 2018 Eleventh International Conference on Contemporary Computing (IC3), pp. 1–5. doi: 10.1109/IC3.2018.8530608. Makkonen, Simo and Risto Lahdelma (2006). „Non-convex power plant modelling in energy optimisation“. In: European Journal of Operational Research 171.3, pp. 1113– 1126. doi:10.1016/j.ejor.2005.01.020. Mancò, Giulia, Umberto Tesio, Elisa Guelpa, and Vittorio Verda (2024). „A review on multi energy systems modelling and optimization“. In: Applied Thermal Engineering 236, p. 121871. doi:10.1016/j.applthermaleng.2023.121871. Maurer, Florian, Felix Nitsch, Johannes Kochems, Christoph Schimeczek, Volker Sander, and Sebastian Lehnhoff (2024). „Know Your Tools - A Comparison of Two Open AgentBased Energy Market Models“. In: 2024 20th International Conference on the European Energy Market (EEM). IEEE, pp. 1–8. doi:10.1109/EEM60825.2024.10609021. Meadows, Donella H, Dennis L Meadows, Jørgen Randers, and William W Behrens (1972). The Limits to Growth. Potomac Associates, Universe Books. Miskiw, Kim K., Emil Kraft, and Stein-Erik Fleten (2025). „Coordinated bidding in sequential electricity markets: Effects of price-making“. In: Energy Economics 144. doi: 10.1016/j.eneco.2025.108316. Mörtenkötter, Hendrik, Hanzhe Xing, Bryn Pickering, Stuart Scott, and Michael Bucksteeg (2025). „Impact of Smaller Market Zones on the Market Value of Battery Storage Systems in Central Europe“. In: doi:10.2139/ssrn.5216465. References 129 F. Nitsch Mudelsee, Manfred (2019). „Trend analysis of climate time series: A review of methods“. In: Earth-Science Reviews 190, pp. 310–322. issn: 0012-8252. doi:10.1016/j. earscirev.2018.12.005. Nado, Zachary et al. (2021). „Uncertainty Baselines: Benchmarks for Uncertainty & Robustness in Deep Learning“. In: doi:10.48550/arXiv.2106.04015. Nakata, Toshihiko, Diego Silva, and Mikhail Rodionov (2011). „Application of energy system models for designing a low-carbon society“. In: Progress in Energy and Combustion Science 37.4, pp. 462–502. doi:10.1016/j.pecs.2010.08.001. Nienhaus, Kristina, Christoph Schimeczek, Ulrich Frey, Evelyn Sperber, Seyedfarzad Sarfarazi, Felix Nitsch, Johannes Kochems, and A. Achraf El Ghazi (2025). AMIRIS Examples.doi:10.5281/zenodo.7789049. Nitsch, Felix (2023a). „Community Building: Modelling electricity markets using the framework FAME and the model AMIRIS“. In: doi:10.5281/zenodo.10544479. Nitsch, Felix (2023b). „focapy: Timeseries forecasting in Python“. In: doi:10 . 5281 / zenodo.7792750. Nitsch, Felix (Mar. 2025a). AMIRIS-CHARGIN. Version v1.0.0. doi:10.5281/zenodo. 15100508. Nitsch, Felix (2025b). PriceForecasterApi Agent & AMIRIS-PriceForecast Extension: Enhanced electricity price forecasts in AMIRIS v3.4.doi:10.5281/zenodo.14963197. Nitsch, Felix, Marc Deissenroth-Uhrig, Christoph Schimeczek, and Valentin Bertsch (2021). „Economic evaluation of battery storage systems bidding on day-ahead and automatic frequency restoration reserves markets“. In: Applied Energy 298, p. 117267. issn: 0306-2619. doi:10.1016/j.apenergy.2021.117267. Nitsch, Felix and A. Achraf El Ghazi (2023). „Energy systems analysis considering crossborder electricity trading: Coupling day-ahead markets in an agent-based electricity market model“. In: doi:10.5281/zenodo.10544676. Nitsch, Felix, Ulrich Frey, and Christoph Schimeczek (2023). „AMIRIS-Scengen: A Scenario Generator for the Open Electricity Market Model AMIRIS“. In: doi:10.5281/ zenodo.8382790. Nitsch, Felix and Christoph Schimeczek (2023). Comparison of electricity price forecasting methods for use in agent-based energy system models.doi:10.5281/zenodo. 14962296. Nitsch, Felix and Christoph Schimeczek (2024). Flexible and accurate price time series forecasts for the application in electricity market simulations.doi:10.5281/zenodo. 10678985. References 130 F. Nitsch Nitsch, Felix and Christoph Schimeczek (2025a). „AMIRIS-PriceForecast: An Extension to the agent-based electricity market model AMIRIS providing external electricity price forecasts“. In: doi:10.5281/zenodo.14907870. Nitsch, Felix and Christoph Schimeczek (2025b). ML-Based Price Forecasts in the Open Electricity Market Model AMIRIS.doi:10.5281/zenodo.14935333. Nitsch, Felix, Christoph Schimeczek, and Valentin Bertsch (2024). „Applying machine learning to electricity price forecasting in simulated energy market scenarios“. In: Energy Reports 12, pp. 5268–5279. issn: 2352-4847. doi:10.1016/j.egyr.2024.11.013. Nitsch, Felix, Christoph Schimeczek, and Valentin Bertsch (2025). Profitability of Competing Flexibility Options in Renewable-Dominated Energy Markets: Combining AgentBased and Machine Learning Approaches. Preprint available at SSRN. doi:10.2139/ ssrn.5320926. Nitsch, Felix, Christoph Schimeczek, Ulrich Frey, and Benjamin Fuchs (2023). „FAME-Io: Configuration tools for complex agent-based simulations“. In: Journal of Open Source Software 8.84, p. 4958. doi:10.21105/joss.04958. Nitsch, Felix, Christoph Schimeczek, Kristina Nienhaus, Ulrich J. Frey, Evelyn Sperber, Johannes Kochems, Aboubakr Achraf El Ghazi, and Leonard Willeke (Mar. 2025). AMIRIS – Lessons from three years of collaborative development. Version v1.0. doi: 10.5281/zenodo.15100373. Nitsch, Felix, Christoph Schimeczek, and Sebastian Wehrle (2021). Back-testing the agentbased model AMIRIS for the Austrian day-ahead electricity market.doi:10.5281/ zenodo.5726737. Nitsch, Felix, Christoph Schimeczek, Manuel Wetzel, Jan Buschmann, and Kai von Krbek (2023). „iog2x - Input/Output Processing of GDX Files“. In: doi:10.5281/zenodo. 8283336. Nitsch, Felix, Yvonne Scholz, et al. (2023). Versorgungsicherheit in Deutschland und Mitteleuropa während Extremwetter-Ereignissen (VERMEER): Der Beitrag des transnationalen Stromhandels bei hohen Anteilen erneuerbarer Energien.url:https://elib. dlr.de/196641/. Nitsch, Felix, Evelyn Sperber, Ulrich J. Frey, Aboubakr Achraf El Ghazi, Johannes Kochems, and Christoph Schimeczek (2025). Extending agent-based simulation capabilities by coupling external models using FastAPI.doi:10.5281/zenodo.14976781. Nitsch, Felix and Manuel Wetzel (2022). „Profitability of Power-to-Heat-to-Power Storages in Scenarios With High Shares of Renewable Energy“. In: Energy Proceedings 28. doi: 10.46855/energy-proceedings-10258. References 131 F. Nitsch Nitsch, Felix, Manuel Wetzel, Hans Christian Gils, and Kristina Nienhaus (2024). „The future role of Carnot batteries in Central Europe: Combining energy system and market perspective“. In: Journal of Energy Storage 85, p. 110959. issn: 2352-152X. doi: 10.1016/j.est.2024.110959. Ölmez, Mehmet Emre, Ibrahim Ari, and Gülfem Tuzkaya (2024). „A comprehensive review of the impacts of energy storage on power markets“. In: Journal of Energy Storage 91, p. 111935. issn: 2352-152X. doi:10.1016/j.est.2024.111935. OpenAI et al. (2024). GPT-4 Technical Report.doi:10.48550/arXiv.2303.08774. Orths, Antje, C. Lindsay Anderson, Tom Brown, Joseph Mulhern, Danny Pudjianto, Bernhard Ernst, Mark O‘Malley, James McCalley, and Goran Strbac (2019). „Flexibility From Energy Systems Integration: Supporting Synergies Among Sectors“. In: IEEE Power and Energy Magazine 17.6, pp. 67–78. doi:10.1109/MPE.2019.2931054. Osman, Ahmed I, Lin Chen, Mingyu Yang, Goodluck Msigwa, Mohamed Farghali, Samer Fawzy, David W Rooney, and Pow-Seng Yap (2023). „Cost, environmental impact, and resilience of renewable energy under a changing climate: a review“. In: Environmental chemistry letters 21.2, pp. 741–764. doi:10.1007/s10311-022-01532-8. Pallonetto, Fabiano, Changhong Jin, and Eleni Mangina (2022). „Forecast electricity demand in commercial building with machine learning models to enable demand response programs“. In: Energy and AI 7, p. 100121. doi:10.1016/j.egyai.2021.100121. Pape, Christian, Simon Hagemann, and Christoph Weber (2016). „Are fundamentals enough? Explaining price variations in the German day-ahead and intraday power market“. In: Energy Economics 54.Supplement C, pp. 376–387. issn: 0140-9883. doi: 10.1016/j.eneco.2015.12.013. Paszke, Adam et al. (2019). „PyTorch: An Imperative Style, High-Performance Deep Learning Library“. In: Advances in Neural Information Processing Systems 32. Curran Associates, Inc., pp. 8024–8035. doi:10.48550/arXiv.1912.01703. Petropoulos, Fotios et al. (2022). „Forecasting: theory and practice“. In: International Journal of Forecasting 38.3, pp. 705–871. issn: 0169-2070. doi:10 . 1016 / j . ijforecast.2021.11.001. Pfenninger, Stefan (2024). „Open code and data are not enough: understandability as design goal for energy system models“. In: Progress in Energy 6.3, p. 033002. doi: 10.1088/2516-1083/ad371e. Pfenninger, Stefan, Adam Hawkes, and James Keirstead (2014). „Energy systems modeling for twenty-first century energy challenges“. In: Renewable and Sustainable Energy Reviews 33, pp. 74–86. issn: 1364-0321. doi:10.1016/j.rser.2014.02.003. References 132 F. Nitsch Pfenninger, Stefan, Lion Hirth, et al. (2018). „Opening the black box of energy modelling: Strategies and lessons learned“. In: Energy Strategy Reviews 19, pp. 63–71. issn: 2211467X. doi:10.1016/j.esr.2017.12.002. Pickard, William F (2011). „The history, present state, and future prospects of underground pumped hydro for massive energy storage“. In: Proceedings of the IEEE 100.2, pp. 473–483. doi:10.1109/JPROC.2011.2126030. Radford, Alec, Jong Wook Kim, Tao Xu, Greg Brockman, Christine Mcleavey, and Ilya Sutskever (2023). „Robust Speech Recognition via Large-Scale Weak Supervision“. In: Proceedings of the 40th International Conference on Machine Learning. Ed. by Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett. Vol. 202. Proceedings of Machine Learning Research. PMLR, pp. 28492–28518. doi:10.48550/arXiv.2212.04356. Rahman, Abidur, Omar Farrok, and Md Mejbaul Haque (2022). „Environmental impact of renewable energy source based electrical power plants: Solar, wind, hydroelectric, biomass, geothermal, tidal, ocean, and osmotic“. In: Renewable and sustainable energy reviews 161, p. 112279. doi:10.1016/j.rser.2022.112279. Ramsebner, Jasmine, Reinhard Haas, Amela Ajanovic, and Martin Wietschel (2021). „The sector coupling concept: A critical review“. In: Wiley interdisciplinary reviews: energy and environment 10.4, e396. doi:10.1002/wene.396. Rasp, Stephan and Sebastian Lerch (2018). „Neural Networks for Postprocessing Ensemble Weather Forecasts“. In: Monthly Weather Review 146.11, pp. 3885–3900. issn: 15200493. doi:10.1175/MWR-D-18-0187.1. Reeg, Matthias (2019). „AMIRIS-ein agentenbasiertes Simulationsmodell zur akteursspezifischen Analyse techno-ökonomischer und soziotechnischer Effekte bei der Strommarktintegration und Refinanzierung erneuerbarer Energien“. PhD thesis. Deutsches Zentrum für Luft-und Raumfahrt, Institut für Technische Thermodynamik. url:https: //nbn-resolving.org/urn:nbn:de:bsz:14-qucosa2-347643. Reza, MS, MA Hannan, Pin Jern Ker, M Mansor, MS Hossain Lipu, MJ Hossain, and TM Indra Mahlia (2023). „Uncertainty parameters of battery energy storage integrated grid and their modeling approaches: A review and future research directions“. In: Journal of Energy Storage 68, p. 107698. issn: 2352-152X. doi:10.1016/j.est.2023.107698. Ringkjøb, Hans-Kristian, Peter M. Haugan, and Ida Marie Solbrekke (2018). „A review of modelling tools for energy and electricity systems with large shares of variable renewables“. In: Renewable and Sustainable Energy Reviews 96, pp. 440–459. issn: 1364-0321. doi:10.1016/j.rser.2018.08.002. References 133 F. Nitsch Ringler, Philipp, Dogan Keles, and Wolf Fichtner (2016). „Agent-based modelling and simulation of smart electricity grids and markets – A literature review“. In: Renewable and Sustainable Energy Reviews 57, pp. 205–215. issn: 1364-0321. doi:10.1016/j. rser.2015.12.169. Rogelj, Joeri, Piers M Forster, Elmar Kriegler, Christopher J Smith, and Roland Séférian (2019). „Estimating and tracking the remaining carbon budget for stringent climate targets“. In: Nature 571.7765, pp. 335–342. doi:10.1038/s41586-019-1368-z. Rombach, Robin, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer (2022). „High-resolution image synthesis with latent diffusion models“. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 10684– 10695. doi:10.48550/arXiv.2112.10752. Ruhnau, Oliver, Michael Bucksteeg, David Ritter, Richard Schmitz, Diana Böttger, Matthias Koch, Arne Pöstges, Michael Wiedmann, and Lion Hirth (2022). „Why electricity market models yield different results: Carbon pricing in a model-comparison experiment“. In: Renewable and Sustainable Energy Reviews 153, p. 111701. issn: 1364-0321. doi:10.1016/j.rser.2021.111701. Ruhnau, Oliver, Clemens Stiewe, Jarusch Muessel, and Lion Hirth (2023). „Natural gas savings in Germany during the 2022 energy crisis“. In: Nature Energy 8.6, pp. 621–628. Sarfarazi, Seyedfarzad, Shima Sasanpour, and Valentin Bertsch (2024). „Integration of energy communities in the electricity market: A hybrid agent-based modeling and bilevel optimization approach“. In: Energy Reports 12, pp. 1178–1196. issn: 2352-4847. doi:10.1016/j.egyr.2024.06.052. Sarfarazi, Seyedfarzad, Shima Sasanpour, and Karl-Kiên Cao (2023). „Improving energy system design with optimization models by quantifying the economic granularity gap: The case of prosumer self-consumption in Germany“. In: Energy Reports 9, pp. 1859– 1874. issn: 2352-4847. doi:10.1016/j.egyr.2022.12.145. Schelling, Thomas C (2006). Micromotives and macrobehavior. WW Norton & Company. Schimeczek, Christoph, Marc Deissenroth-Uhrig, Ulrich Frey, Benjamin Fuchs, A. Achraf El Ghazi, Manuel Wetzel, and Kristina Nienhaus (2023). „FAME-Core: An open Framework for distributed Agent-based Modelling of Energy systems“. In: Journal of Open Source Software 8.84, p. 5087. doi:10.21105/joss.05087. Schimeczek, Christoph, Kristina Nienhaus, Ulrich Frey, Evelyn Sperber, Seyedfarzad Sarfarazi, Felix Nitsch, Johannes Kochems, and A. Achraf El Ghazi (2023). „AMIRIS: Agent-based Market model for the Investigation of Renewable and Integrated energy Systems“. In: Journal of Open Source Software 8.84, p. 5041. doi:10.21105/joss. 05041. References 134 F. Nitsch Schimeczek, Christoph, Felix Nitsch, Marc Deissenroth-Uhrig, Ulrich Frey, Benjamin Fuchs, A. Achraf El Ghazi, Manuel Wetzel, and Kristina Nienhaus (2020). „FAME: Open Framework for distributed Agent-based Modelling of Energy systems“. In: url: https://gitlab.com/fame-framework. Schimeczek, Christoph, Manish Khanra, and Tim Signer (2025). Modelling Flexibility Options in Agent Based Models.doi:10.5281/zenodo.15172712. Schimeczek, Christoph and Felix Nitsch (2024). AMIRIS-Py.doi:10. 5281/ zenodo. 14273248. Schlecht, Ingmar, Christoph Maurer, and Lion Hirth (2024). „Financial contracts for differences: The problems with conventional CfDs in electricity markets and how forward contracts can help solve them“. In: Energy Policy 186, p. 113981. issn: 0301-4215. doi: 10.1016/j.enpol.2024.113981. Schleussner, Carl-Friedrich et al. (2016). „Science and policy characteristics of the Paris Agreement temperature goal“. In: Nature climate change 6.9, pp. 827–835. doi:10. 1038/nclimate3096. Schmidt, Oliver and Iain Staffell (Sept. 2023). Monetizing Energy Storage: A Toolkit to Assess Future Cost and Value. Oxford University Press. isbn: 9780192888174. doi: 10.1093/oso/9780192888174.001.0001. Schulz, Benedikt and Sebastian Lerch (Jan. 2022). „Machine Learning Methods for Postprocessing Ensemble Forecasts of Wind Gusts: A Systematic Comparison“. In: Monthly Weather Review 150.1, pp. 235–257. issn: 1520-0493. doi:10 . 1175 / mwr - d - 21 - 0150.1. Siala, Kais, Mathias Mier, Lukas Schmidt, Laura Torralba-Díaz, Siamak Sheykhha, and Georgios Savvidis (2022). „Which model features matter? An experimental approach to evaluate power market modeling choices“. In: Energy 245, p. 123301. issn: 03605442. doi:10.1016/j.energy.2022.123301. Signer, Tim, Jonathan Mack, Max Kleinebrahm, Viktor Slednev, Thorsten Weiskopf, and Wolf Fichtner (2025). „Integration of Block, Linked and Loop Bids in a Market Coupling Algorithm“. In: 2025 21st International Conference on the European Energy Market (EEM). IEEE, pp. 1–7. doi:10.1109/EEM64765.2025.11050140. Silva-Rodriguez, Lina, Anibal Sanjab, Elena Fumagalli, Ana Virag, and Madeleine Gibescu (2022). „Short term wholesale electricity market designs: A review of identified challenges and promising solutions“. In: Renewable and Sustainable Energy Reviews 160, p. 112228. issn: 1364-0321. doi:10.1016/j.rser.2022.112228. Song, Hui, Chen Liu, Ali Moradi Amani, Mingchen Gu, Mahdi Jalili, Lasantha Meegahapola, Xinghuo Yu, and George Dickeson (2024). „Smart optimization in battery energy References 135 F. Nitsch storage systems: An overview“. In: Energy and AI 17, p. 100378. issn: 2666-5468. doi: 10.1016/j.egyai.2024.100378. Sorourifar, Farshud, Victor M. Zavala, and Alexander W. Dowling (2020). „Integrated Multiscale Design, Market Participation, and Replacement Strategies for Battery Energy Storage Systems“. In: IEEE Transactions on Sustainable Energy 11.1, pp. 84–92. doi:10.1109/TSTE.2018.2884317. Sperber, Evelyn, Christoph Schimeczek, Ulrich Frey, Karl Kiên Cao, and Valentin Bertsch (2025). „Aligning heat pump operation with market signals: A win-win scenario for the electricity market and its actors?“ In: Energy Reports 13, pp. 491–513. issn: 2352-4847. doi:10.1016/j.egyr.2024.12.028. Staffell, Iain and Stefan Pfenninger (2018). „The increasing impact of weather on electricity supply and demand“. In: Energy 145, pp. 65–78. issn: 0360-5442. doi:10.1016/ j.energy.2017.12.051. Strbac, Goran et al. (2021). „Decarbonization of electricity systems in Europe: Market design challenges“. In: IEEE Power and Energy Magazine 19.1, pp. 53–63. doi:10. 1109/MPE.2020.3033397. Stute, Judith and Marian Klobasa (2024). „How do dynamic electricity tariffs and different grid charge designs interact? - Implications for residential consumers and grid reinforcement requirements“. In: Energy Policy 189, p. 114062. issn: 0301-4215. doi: 10.1016/j.enpol.2024.114062. Tesfatsion, Leigh (2006). „Chapter 16 Agent-Based Computational Economics: A Constructive Approach to Economic Theory“. In: ed. by Leigh Tesfatsion and K.L. Judd. Vol. 2. Handbook of Computational Economics. Elsevier, pp. 831–880. doi:10.1016/ S1574-0021(05)02016-2. Torralba-Diaz, Laura, Christoph Schimeczek, Matthias Reeg, Georgios Savvidis, Marc Deissenroth-Uhrig, Felix Guthoff, Benjamin Fleischer, and Kai Hufendiek (2020). „Identification of the Efficiency Gap by Coupling a Fundamental Electricity Market Model and an Agent-Based Simulation Model“. In: Energies 13.15, p. 3920. issn: 1996-1073. doi:10.3390/en13153920. Torralba-Díaz, Laura, Christoph Schimeczek, Johannes Kochems, and Kai Hufendiek (2024). „Iterative coupling of a fundamental electricity market model and an agentbased simulation model to reduce the efficiency gap“. In: Energy 310, p. 133192. issn: 0360-5442. doi:10.1016/j.energy.2024.133192. Twidell, John (2021). Renewable energy resources. Routledge. doi:10 . 4324 / 9780429452161. References 136 F. Nitsch Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin (2023). Attention Is All You Need.doi: 10.48550/arXiv.1706.03762. Walser, Thilo and Alexander Sauer (2021). „Typical load profile-supported convolutional neural network for short-term load forecasting in the industrial sector“. In: Energy and AI 5, p. 100104. doi:10.1016/j.egyai.2021.100104. Ward, KR, Richard Green, and Iain Staffell (2019). „Getting prices right in structural electricity market models“. In: Energy Policy 129, pp. 1190–1206. issn: 0301-4215. doi:10.1016/j.enpol.2019.01.077. Weidlich, Anke and Daniel Veit (2008). „A critical survey of agent-based wholesale electricity market models“. In: Energy Economics 30.4, pp. 1728–1759. issn: 0140-9883. doi:10.1016/j.eneco.2008.01.003. Weiskopf, Thorsten, Tim Signer, Jonathan Stelzer, Armin Ardone, and Wolf Fichtner (2024). PowerACE from a computational view. 10th bwHPC Symposium (2024), Freiburg im Breisgau, Deutschland, 25.–26. September 2024. doi:10 . 5445 / IR / 1000174620. Wetzel, Manuel et al. (2024). „REMix: A GAMS-based framework for optimizing energy system models“. In: Journal of Open Source Software 9.99, p. 6330. doi:10.21105/ joss.06330. Willeke, Leonard, Johannes Kochems, and Kristina Nienhaus (2025). „Agent-Based Modelling of Investment Decisions in the Electricity Sector“. In: 2025 21st International Conference on the European Energy Market (EEM). IEEE, pp. 1–7. doi:10.1109/ EEM64765.2025.11050320. Yourdon, Edward and Larry L. Constantine (1979). Structured design. Fundamentals of a discipline of computer program and systems design. YOURDON press. isbn: 0138544719. Zhou, Zhi-Hua (2021). Machine learning. Springer nature. isbn: 9811519668. doi:10. 1007/978-981-15-1967-3. Zöphel, Christoph, Steffi Schreiber, Theresa Müller, and Dominik Möst (2018). „Which flexibility options facilitate the integration of intermittent renewable energy sources in electricity systems?“ In: Current Sustainable/Renewable Energy Reports 5, pp. 37–44. doi:10.1007/s40518-018-0092-x. References 137 A Supplementary peer-reviewed publications A.1 AMIRIS: Agent-based Market model for the Investigation of Renewable and Integrated energy Systems Authors: Christoph Schimeczek, Kristina Nienhaus, Ulrich Frey, Evelyn Sperber, Seyedfarzad Sarfarazi, Felix Nitsch, Johannes Kochems, A. Achraf El Ghazi Corresponding Author: Christoph Schimeczek Journal: Journal of Open Source Software Volume: 8 (84) DOI: 10.21105/joss.05041 Status: Published 17 April 2023 Licence: Open Access, CC BY 4.0 Abstract: AMIRIS is an agent-based model (ABM) to simulate electricity markets. The focus of this bottom-up model is on the business-oriented decisions of actors in the energy system. These actors are represented as prototypical agents in the model, each with own complex decision-making strategies. The bidding decisions are based on the assessment of electricity market prices and generation forecasts, and diverse actors deciding on different time scales may be modelled. In particular, the agents’ behaviour does not only reflect marginal prices, but can also consider effects of support instruments like market premia, uncertainties and limited information, or market power. This allows assessing which policy or market design is best suited to an economic and effective energy system. The simulations generate results on the dispatch of power plants and flexibility options, technology-specific market values, development of system costs or CO2emissions. One important output of the model are simulated market prices. AMIRIS is developed in Java using the FAME-Coreframework (Schimeczek et al., 2023) and is available on GitLab. One important design goal was to make assumptions and calculations as transparent as possible in order to improve reproducibility. AMIRIS was successfully tested on different computer systems, ranging from desktop PCs to high performance clusters. 138