Control of aggregated virtual synchronous generators for PV plants considering communication delays

6512 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 60, NO. 4, JULY/AUGUST 2024 Control of Aggregated Virtual Synchronous Generators for PV Plants Considering Communication Delays Daniel del Rivero , Graduate Student Member, IEEE, Pablo García , Senior Member, IEEE, Cristian Blanco , Senior Member, IEEE, and Ángel Navarro-Rodríguez , Member, IEEE Abstract—In this paper, a new method for the delay compensation when using an aggregation of virtual synchronous generators is proposed. Lack of inertia in power converters can potentially provoke stability issues that can be mitigated by the use of virtual inertia techniques. Among those, the Virtual Synchronous Generator(VSG)concepthasreceivedstrongimpulseinthelastyears.This paper is focused on the idea of using the distributed VSG concept in a renewable power plant, in which a single Synchronous Central Angle Controller (SCAC) is used for the power control exchange at the Point of Connection (PoC), while distribution control units are employedforthe localinvertercontrol.This idea, alreadydiscussed in the literature, is in here extended to consider the implementation on industrial string-level commercial power converters, recalling theimportanceofaccessiblemeasurementsandcommunicationdelays.In order tovalidatethe proposal,firstly communicationdelays are measured and modelled. Following, simulations with different SCAC operating modes are conducted, and finally experimental results validation of different operation modes with commercial converters are presented. Index Terms—Communication delay, real-time simulation, smith predictor, virtual synchronous generator. I. INTRODUCTION THE world’s power generation is currently moving toward a more sustainable and environmentally friendly approach. This is due to the usage of Distributed Energy Generation (DEG) facilities based on Renewable Energy Sources (RES) has replaced fossil fuels because of their significant environmental Manuscript received 12 September 2023; revised 25 January 2024; accepted 26 February 2024. Date of publication 13 March 2024; date of current version 22 July 2024. Paper 2023-SECSC-1316.R1, presented at the 2022 IEEE Energy Conversion Congress and Exposition, Detroit, MI, USA, Oct. 09–13, and approved for publication in the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Renewable and Sustainable Energy Conversion Systems Committee of the IEEE Industry Applications Society [DOI:10.1109/ECCE50734.2022.9947466]. This work was supported in part by European Union’s H2020 Research and Innovation Program under Grant 864459 (UE-19TALENT-864459), in part by the Spanish Ministry of Innovation and Science under Grant CI-MCINN-23-PID2022-139479OB-C22 and Grant MCINN-22TED2021-129796B-C21, and in part by the Principality of Asturias, FICYT, FEDER Funds under Grant SV-PA-21-AYUD/2021/57546. (Corresponding author: Daniel del Rivero.) The authors are with the Department Electrical Engineering, University of Oviedo, 33203 Gijón, Spain (e-mail: [email protected]; garciafpablo @uniovi.es; [email protected]; nav[email protected]). Color versions of one or more figures in this article are available at https://doi.org/10.1109/TIA.2024.3377169. Digital Object Identifier 10.1109/TIA.2024.3377169 cost (greenhouse emissions, lack of source material, etc.). Additionally, the most widely used RES, like photovoltaic (PV) and wind power, are becoming more affordable, offering improved Levelized Cost of Electricity (LCOE) indices [1]. However, the inclusion of this kind of generation systems provokesa weakerpowersystem dueto the inertia reduction currently provided by synchronous generators with rotating mass, to a power converter-based system with little to no inertia [1], [2],[3]. Since power converters lack both inertia and damping, this problem could affect the power grid’s stability. It is currently understood that grid-forming and grid-supporting services must be taken into account in the design when significant penetration of DEG, with aggregated sizes comparable to traditional power plants [4],[5],[6]. This is where the Virtual Synchronous Generator (VSG) approach arises. For the power electronics-based DEG/RES units, this control method enables the emulation of the dynamic characteristics of a real or arbitrary Synchronous Generator (SG) [2],[3]. Using Energy Storage Systems (ESS), power converters, and an appropriate control mechanism, the virtual inertia concept is applied to provide a specific amount of inertia for supplied (usually in the range of ms to a few dozen seconds). In this way, VSG establishes the framework for later widespread application in RES systems without jeopardizing system stability. Several VSG approaches have been explored and implemented [1], [2],[3],[4],[5],[6],[7],[8],[9],[10],[11]. Two alternative implementations exist. A first approach refers to those methods rooted in mathematical equations (e.g., synchronverters [8],[9], Kawasaki Heavy Industries [10], VISMA and IEPE topologies [9]). A second group relies on swing equations (e.g., Ise Lab’s topology [11], the Synchronous Power Controller [5], [12], Virtual Oscillator Control [9]). The Synchronous Power Controller (SPC) is a prevalent topology for virtual inertia implementation, synthesizing the electromechanical and electrical characteristics of a SG. This approach regulates inverter frequency by employing virtual inertia and damping factors to counteract grid frequency disturbances. SPC operates with inner current and outer voltage control loops, using a virtual admittance to establish a cascaded control loop. SPC is typically integrated into the local control of each inverter, offering frequency and voltage support at local PV collector connection points [5]. However, for PV plants, © 2024 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ RIVERO et al.: CONTROL OF AGGREGATED VIRTUAL SYNCHRONOUS GENERATORS FOR PV PLANTS CONSIDERING COMMUNICATION DELAYS 6513 Fig. 1. (a) SCAC scheme for n-converters. (b) Simplified Local control scheme. The light green control loop is implemented in the central controller. The orange control loop is implemented in the local controller [13]. (c) Added AVG Control for grid forming capabilities [2]. grid support is ideally expected at the Point of Connection (PoC). With that motivation, a modification of the SPC designed to provide grid support at the PoC has been proposed, the Synchronous Central Angle Controller (SCAC) [4]. The SCAC technique suggests simultaneously driving several converters, emulating a unique SG in the PoC, giving rise to an aggregated VSG. This concept presents the idea of a single virtual rotor, emulated at the PoC, where the electromechanical model of the SG is considered (central control architecture). Hence, the SG inertia and damping response are emulated at the PoC. Regardless of the distances between the local converters, this control structure enables the system operator at the PoC to control the exchange of power (both active and reactive), allowing each converter to distribute the energy to be delivered under different operating modes (power and frequency support) on their own. A more detailed explanation can be found in the literature, where both, central and local control systems are detailed [7]. A concept for a structure with Nconverters is shown in Fig. 1, where a central controller handles the local controller references of each converter. In this system, it is possible to independently control the exchanged active and reactive power, asindicated inthe dynamic model and controlloops. InFig. 1(a), a comprehensive connection diagram illustrates the interconnection of various converters. In Fig. 1(b), the control block diagram is depicted, with the global controller in green and the local controller for each power converter in orange. Each power converter requires a replicated local controller tailored to its specific characteristics. Further explanations for these blocks are provided below. The key control system is implemented in the central controller, where the inertia is emulated by the swing equation of a virtual synchronous generator (see (1) and (2)). dδsm dt=Δωr(1) JdΔωr dt=Pm−Pe ωB|−DΔωr=ΔP ωB −DΔωr(2) where δsm is the power angle, Δωris the angular speed deviation of the rotor, Jis the SG inertia, Pmis the mechanical power, Peis the electrical power, Dis the damping constant and ωBis the base frequency. In [5], the electromechanical control has been studied, and a frequency analysis has been taken into account to obtain the power loop control HM(3).kp,ki, and KD have been designed, according to the required inertia constant and frequency droop slope, respectively. HM=Δωr ΔP =kp+kis s+kD ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ kp=ω2 n Pmax kD=ω2 nD Pmax ki=2Pmax ξωn−ω2 nD P2 max wn=Pmax ωB 2HSN H=Jω2 B 2SN (3) In (3),His the inertia constant, Pmax is the maximum active power of the converter, SNis the nominal power, ωnis the natural frequency and ξis the damping factor. In this control system, the dynamic response is mainly supported by the inertia, while the frequency droop supports the steady state behavior. Based on the analysis presented in [4],[5],(4) shows the relationship between ΔPoand the frequency change Δωg. ΔPo Δωg =SN ωBDP (4) The internal time-domain variables and control loops (current/voltage) of the converter are presumed to be accessible by this control system, though. However only active and reactive 6514 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 60, NO. 4, JULY/AUGUST 2024 power set-points are normally externally accessible for commercial converters that have already been installed, typically via a communications link. So, a modification of this control is required for a wider applicability. Due to its wide adoption as the go-to solution in power plants, MODBUS TCP is proposed as the communication system [14]. In the following sections, the control system architecture is modified in accordance with the prior motivation using a communication-based implementation. In order to perform that implementation, the system communication delay is measured, the integration of the Smith predictor (SP) in the system is explained, and also a small stability analysis is performed. For validating the model, several working operations will be tested in local simulation and in real-time operation with commercial converters. Those working modes are: I) Active and reactive injection. II) Power support operation, taking into account grid operator requests. III) Frequency support operation by power management depending on frequency variations. IV) Phasejump in the grid voltage performance. V) Island operation. First, a local simulation is used to evaluate the control system, and following a real-time hardware controller with commercial converters is employed for the first three cases. The main article contribution is the proposal of using a distributed virtual synchronous generator in an industrial environment considering communication delays and their compensation. This paper is based on the paper in [13] by the same authors, with extended analysis and results. The added content includes a stability analysis of the system considering the delay impact. Regarding the results, a variety of different operating conditions is included, considering Hardware-In-the-Loop (HIL) and Power-Hardware-In-the-Loop (PHIL) validation schemes. II. PROPOSED CONTROL SYSTEM Most of the VSG techniques integrate their controllers into each converter’s firmware as an add-on. However, as it was already indicated, the SCAC technique requires having access to different control actions and sensor readings (current, voltages). This paper proposes a new control structure that can be applied to already-existing commercial converters, that only requires access to active and reactive powers set-points and measurements obtained by MODBUS TCP communications and the dictionary variables included in the SunSpec DER specification [15].The method does not require any additional measuring elements, such as extra voltage and current sensors, which would make implementation more complicated and expensive. Instead, it relies on the RMS voltage, frequency, and active and reactive power communication-based readings from each converter. Therefore, this approach is an appealing solution for the standardization of the VSG concept for a massive implementation in future and existing DEG’s. However, it is important to acknowledge certain limitations when comparing the external implementation of the VSG concept in power converters to its internal counterpart. The external implementation focuses solely on the fundamental component for signal reconstruction in the time domain. Additionally, achieving a rapid response is constrained by communication delays and the necessary time for reconstruction, which Fig. 2. (a) MODBUS TCP frequency and RMS voltage. (b) Instantaneous values from the RMS voltage, frequency, and clock signal obtained by a VCO (Voltage Controller Oscillator) [16]. (c) Active and reactive reference power calculation [13]. entails one fundamental cycle, taking into account RMS values, and it is feasible only within the internal control of the power converter. For the application of this concept, instantaneous voltage signals for each converter and PoC are built from the RMS values as shown in Fig. 2(a) and (b). These signals are employed in the control system’s feedback variables (as a replacement of bold and circled variables in the original scheme shown in Fig. 1(b). At the same time, as it was already indicated, commercial converters typically accept power (active/reactive) setpoints. However, for the implementation of the VSG concept, currentcommandscomputedfrom the control systemareneeded instead. This proposal derives the power references from the current references and the reconstructed voltage signals, as shown in Fig. 2in the α−βreference frame. III. DELAY COMPENSATION Each converter control unit communicates with the central controller in the proposed renewable energy plant application using MODBUS TCP. The communication between the central controller and each of the distributed units may experience some delay since MODBUS TCP is not a real-time protocol, the delay depending on the number of components in the bus and the distance. The effectiveness of the closed-loop system is compromised by these delays, which have a direct impact on the control instructions transmitted from the central controller and the provided feedback information. Considering that MODBUS TCP is not a real-time protocol, it is expected a variable delay distribution. Accordingly, the delay statistical distribution is modeled in this section and a compensating mechanism is discussed. A. Delay Modelling In the literature, there are several proposals to model random delays. One of them uses the Markov chain [17],[18] as its foundation. A stochastic model called a Markov chain discretely represents certain potential states or events (in this case delays). RIVERO et al.: CONTROL OF AGGREGATED VIRTUAL SYNCHRONOUS GENERATORS FOR PV PLANTS CONSIDERING COMMUNICATION DELAYS 6515 Fig. 3. Setup for HIL and PIL experimental tests. The control design is made on Matlab and executed in the real-time Speedgoat target. The converters are controlled by writing/reading published MODBUS/TCP variables [13]. The probability of those potential outcomes is solely dependent on the outcome of the prior event. For this paper, to model the communication latency in the local network where the tests are going to be performed, the communication delays in a link using MODBUS TCP protocol are measured. A 30 kW bidirectional dc/dc converter (CNG) from Cinergia SL, similar to the one in Fig. 3, was employed as the PHIL system. This power supply features different emulation units, including batteries and solar panels, and may work as a voltage, current, and power source. Two of these power sources will be used in Section IV for the experimental validation. For performing the measurements, a HIL system (Speedgoat target machine) is used to act as a communication gateway between the converters and the control system in a real-time simulation. These components are also displayed in Fig. 3. The delay measurement procedure is as follows: a digital square reference signal of 0.25 Hz has been supplied simultaneously to the current converter set-point, so it can be used as a trigger signal in an external scope that also captures the output current response. Both signals can be observed in Fig. 4(a).The delay distribution varies between a much wider range, as it can be seen in Fig. 4(b), with a mean value of around 68 ms and a mode of 45 ms. In Fig. 4(c), the time variation of the delay during all the experiments can be easily appreciated. The time variation of the delay throughout the entire experiment is seen in Fig. 4(c). For the delay modelling, a Poisson distribution with the form (5) is chosen, as proposed in queuing theory delay models for communication networks [19],[20]. The Poisson distribution is obtained with the delay evolution from Fig. 4(b), with 795 number of events (k) and the mean (λ) value of 68 ms. That distribution is used as delay estimation for compensating the delays in the SP loop. For the real communication delay, the measured data is used. f(k,λ)=Pr(X=k)=λke−λ k!;λ>0;k=0,1,2... (5) By using the Poisson distribution, the distribution from Fig. 4(d) is obtained. B. Delay Compensation. The Smith Predictor Various methods for delay compensation have been explored, including the study of the Smith predictor (SP) and its modifications [21],[22], the investigation of the Scattering transformation [23],[24], the examination of the linear predictor [25], Fig. 4. (a) Instantaneous delay measurement in the lab (PWM is the sent reference signal and I is the actual current the converter develops). (b) Time variation of delay. (c) Delay histogram from experimental tests. (d) Probability density function of the measured communication delay (M.D) in comparison to Poisson distribution (P.D) with λof 68 ms. [26], and the consideration of predictive control [27],[28], among other strategies. The Scattering transformation serves as amethodtopassivatethecontrolsystem,mitigatingdelayeffects and contributing to stabilization. Similarly, the linear predictor, acommonly usedmodel-free scheme, employs the linear extrapolation concept to predict future control variables. However, for the purposes of this paper, the SP has been chosen due to its simplicity and reliable operation [21]. Ongoing research in this field aims to identify alternative delay compensation methods more suitable for the stochastic nature of communication-based delays. The SP achieves the removal of the delay component from the system’s control loop by incorporating a model of the 6516 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 60, NO. 4, JULY/AUGUST 2024 Fig. 5. (a) Global SP architecture for n-converters (light green block for global control and orange for local control) [17]. (b) Local model for each converter (blue blocks) emulated in the global controller. (c) Basic Smith predictor structure. delay structure, along with a relatively precise modeling of the system plant [22]. In the SCAC system, the delay is presented in the control actions sent to the local converters. The SP compensates for the plant delay, through a plant model ( Gp) and an estimated delay ( etps). The plant model shall be the one between the control actions (δm,ΔE) and the output (Pout,Qout)(see Fig. 5(a). Unfortunately, finding the transfer model for the proposed communications-based control system is not an easy task. The suggested approach is to run a replica of the local control for each converter in the central controller. As many replicas of the local control will be used as there are SCAC converters. It should also be clarified that the local control structure of each of the converters is the same in all of them, simply changing the value of the instantaneous values of each converter. Fig. 5(c) shows the simulated plant for SP which is considered as the affected plant by the delay. Besides, δmand eioutputs of the emulated local model are used for computing the output power using the power (6). Finally, the error between the predicted active and reactive power and the values given by the local control units is used to compensate for the delay (see Fig. 5). Pi=EiVi Xi sin(δm);Qi=Vi Xi (Ecos(δm)−Vg)(6) In order to demonstrate and validate the operation of the SP, the limit stability constant delay is firstly considered while the compensation method is applied. The system used for applying the delay with SP is the one shown in Fig. 5, and without compensation the one shown in Fig. 1. To clarify this point, Fig. 6shows the difference in behavior whenusingornottheSPunderthelimitstabilitydelaycondition. As it can be seen, by setting the limit delay (75 ms), the response with (dLSP ) and without SP (dLNSP ) are clearly different, where the additional overshot created by the delay is mostly removed by applying the SP. It is also included a larger delay (d=80 ms) to illustrate the instability condition above a certain delay level. In this case the unstable case is scaled for representation purposes. In the next subsection, a concise stability analysis is undertaken Fig. 6. System step response with a constant communication delay of d3= 75 ms (which is the stable limit regarding the delay), with (dLSP ) and without delay compensation (dLNSP ). Also, the unstable case (dUNS) is represented for a case of d2=80 ms. For representation, the unstable power is corrected by a factor of 0.005. Ris the reference signal. to complement the explanation and validate the importance of implementing a delay compensation method. C. Stability Analysis In this section, a comparative stability study is conducted, considering both the system without considering communication delays and those that include them, leading to the system instability. This analysis is visually presented in Fig. 7, illustrating Bode diagrams for the different enumerated cases. The stability response will be compared in two different scenarios:onewherethesystemisoperatedwithoutcommunication delays (d1=0 ms) and another where communication delay is introduced at the stability boundary (d2=75 ms). Those scenarios are introduced in Fig. 7, where Bode diagrams are depicted. Initially, the Bode diagram without communication delays was approximated using the Frequency Response Function (FRF) method, a frequency-based measurement function. It consists in a frequency-based measurement function that expresses the frequency domain relationship between an input and output of a system [29]. In the Fig. 7, just the case without delay is included with this method. However, considering that the response at low frequencies closely resembles the transfer function obtained from the system in Fig. 8(due to the inertia control system being slow and the rest being fast at low frequencies), the latter has been employed for the subsequent stability tests. The three first cases were analysed in the Bode by using the system from Fig. 8. RIVERO et al.: CONTROL OF AGGREGATED VIRTUAL SYNCHRONOUS GENERATORS FOR PV PLANTS CONSIDERING COMMUNICATION DELAYS 6517 TABLE I STABILITY VALUES FOR THE DIFFERENT CASES FROM FIG.9 Fig. 7. Bode diagram of the SCAC system considering no delay (d1=0ms), the limit delay which makes unstable the system (d2=75 ms), and the same delaybutcompensatedwith theSPmethod(SP).Furthermorethe Bodediagram extracted by FRF method is also presented (just for non delay case). (a) The amplitude Bode is presented. (b) Shows the phase evolution for the different cases, including the PMs. Notice the low-frequency ranges of the system (x-axis) due to the emulated system inertia (10 s). Table Ishows the stability values of these cases. Fig. 8. Simplified model of SCAC with one converter. As observed in the Bode diagram of Fig. 7and the data presented in Table I, system stability is evident in the absence of communication delays, with a Phase Margin (PM) of 143.44 and an infinite Gain Margin (GM). When the limiting delay (d2) is introduced, the system is positioned at the stability boundary, featuring a GM of 1.001 and a PM of 0.0038. Additionally, upon the introduction of the SP, the system regains its stability margin, displaying a GM of 6.86 and a PM of 97.7. Taking advantage of the stability analysis conducted in the baseline case, a brief assessment of the system’s stability sensitivity has been carried out. Critical parameters such as inertia (H), damping (τ), and droop slope (Kp) were varied across three different scenarios: A) without delay, B) with a limiting delay, and C)limiting delay butemployingSPas acompensation method. These variations are reflected in Fig. 9. Furthermore, the aim of this analysis is to emphasize the significance of certain elements in the control system, demonstrating how they influence the variation of stability margins. Leveraging Fig. 9and Table I, it can be observed that, in the case A) without delays, the modification of Hvalues a) causes the system to become more underdamped but faster as its value decreases. Increasing the value of τb) results in a more overdamped and slower system, while the variation of Kp c) mainly affect to the position of the zeros, moving the root locus to the right, as Kpis increasing. In case B), the system behaves similarly, but with eigenvalues shifted to the right. It is even noticeable that, by increasing Hand decreasing τ,the dominant poles can lead the system to the stability margin, as detailed in Table I. In case C), after delay compensation with SP, the significant eigenvalues return to the negative semi-axis, ensuring system stability. IV. RESULTS In this section some results are presented to validate the proposed compensation method, presenting different working modes of the system. Those operations are tested in both local Simulinksimulations and real-time experimentalproofs through Speedgoat emulator. Real-time tests are based on Fig. 3, where two 30 kW bidirectional dc/dc converters (CNG) from Cinergia S.L are used. In this case, CNG-2 has three strings working as power sources to emulate the power demand from the control system, which will receive the commands from the simulation (Pref1,Qref1,Pref2,Qref2). Those setpoints are sent and written in CNG-2 through MODBUS TCP. The energy computed by the control system will be obtained from CNG-1, which works as a battery emulator in each string, which is running in battery emulation mode to replicate the SCAC idea. As it was stated in Section III, a HIL system is used for real-time simulation. For the case of communication delays between the central controller and the local units, the same variable time delay used in the local simulations (computed in Section III) are used for experimentaltests.However,duetotheCNGconverter’sinternal delay in the processing of the power references and integration windows used for the calculation of the active and reactive power, additional delays are added to the control system (20 ms for active power and 400 ms for reactive power). These delays are also included in the model used by the Smith predictor to achieve better results. This is a critical step, as the Smith predictor will also tackle the additional delays present in a real implementation. For these tests, grid, and VSG models are taken from [4], where the SCAC idea was first published. In this case, the model includes three DEGs connected to a grid and considers a battery locally connected per converter, which is the element that provides/absorbs energy for frequency support. In this case, 6518 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 60, NO. 4, JULY/AUGUST 2024 Fig. 9. Root locus of control system. Base case parameters shown in Table II. (a) Z-P map without delay. (b) Z-P map with delay. (c) Z-P map with SP compensation delay. In the first row of the figure (a), the inertia (H) is varied. In (b) the damping (τ), and in (c) the droop slope (Kp). In Table Ithe improved values are bold. TABLE II SCAC PARAMETERS AND SET-POINTS FOR THE SIMULATION feeder impedance is considered, demonstrating that the system could work in a real implementation with real feeders. Fig. 10 shows an overview of the different tests that are going to be evaluated, showing several operation modes to be validated. Going from a simple active and reactive power reference tracking to a power support operation mode, depending on the grid operator requirements modifying the power exchange with the grid. Besides frequency support capability, by injecting/absorbing energy through an ESS, validating the disturbance rejection capability, which is one of the main purposes of the system. Also the phase-jump reaction of the system and the islanding mode operation are analysed in this paper, to show other extra operations of the system. The cases I, II and III can be seen in Fig. 11, but in terms of power signals in the form of a complete simulation. Fig. 11(a) shows I, II, and III working modes in terms of active power. Fig. 11(b) depicts the behavior in Case I, injecting the required power by the global controller. Fig. 11(c), depicts Case II, for active power management. Besides, Fig. 11(d) and (e) show Case III, which is the injected power when a frequency drop appears in the grid, trying to reduce the frequency variation. These working operations will be explained in more detail in the following subsections. Worth noting that power is oddly shared among the three converters in the next section, providing a distribution showninFig.12,foreachDEG.Simulationparameters aregiven in Table II. Previous simulation is performed by using an ideal grid. However, in order to validate and demonstrate how the SCAC system works, the following simulations will be performed by using a weak grid formed by a simple synchronous generator (with real frequency variations), with a limited power (300 kW) and inertia (0.116 kgm2). In these simulations, the five cases from Fig. 10 are validated: I) active and reactive power injection, II) supporting frequency changes, III) grid operator active power reference tracking, IV) phase angle jump and V) islanding operation. A. Case I Considering what is accounted in Figs. 10(b) and 11, active and reactive power setpoints are established in order to control the power exchange with the grid. This principle is the basis for the other two working modes, showing how the SCAC can manage the required power by the global controller. This operation is tested by a simple local simulation, and also by a real-time simulation. Besides, these power injections are varied by adjusting the virtual admittance in control as Fig. 12 shows, where the unit admittance is divided in 3, and it varies during time. 1) Local Simulation: As it can be seen in Fig. 13(a),the power setpoint is reached, sharing the energy among the DEGs, regarding the power distribution between the DEGs shown in Fig. 12. The same happens with the reactive power in b). 2) Real-Time Experimental Test: As it was aforementioned, the control system is tested in real-time by using the Speedgoat simulator and CNG converters. The experimental results are presented in Fig. 14. As it can be seen active and reactive power are tracked perfectly, quite similar to local results. It is important to note that there is a different delay between active and reactive power, which has to do with the integration window from each variable in the power converter used in the HIL system [13].In case of c), the different steps that appear in the read power are directly the delay of the integration window for active power. Nevertheless, those delays are tackled by the Smith predictor making the system controllable and stable. The difference in the ripple between the local simulation and the real-time is because RIVERO et al.: CONTROL OF AGGREGATED VIRTUAL SYNCHRONOUS GENERATORS FOR PV PLANTS CONSIDERING COMMUNICATION DELAYS 6519 Fig. 10. (a) Power system scheme with three DEGs, showing global and local controllers, battery (SCAC) and PV panels. (b) Working mode I: Active and reactive power setpoints by each DEG, controlled by the SCAC. (c) Working mode II: Power support operation, taking into account grid operator requests. (d) Working mode III: Frequency support operation by power management depending on frequency variations. (e) Working mode IV: Voltage angle stepchange. (f) Working mode V: Islanding mode. the signals read by MODBUS TCP do not have the ripple data. Although the power with ripple is sent, the converter used for the experimental tests filters the component. B. Case II This case aims to demonstrate the main operation mode of the SCAC system. As it was above-mentioned, SCAC system Fig. 11. (a) Active power management for all working modes in each DEG, taking into account the power-sharing between them. Legend G1is the first power converter, G2is the second power converter, G3is the third power converter, DEG is the total power injected by the power plant and Ris the power reference. Case I shows the Active and reactive power setpoints by each generator, controlled by the SCAC. Case II shows the power support operation. Case III shows the frequency of support operation. (b) Zoom of case I for active power injection. (c) Zoom of Case II for active power management. (d) Grid frequency variation. (e) Zoom of Case III for Battery power injection for compensating frequency change. Fig. 12. Virtual admittance variation for Case I and Case II from this section, in order to modify the output power of each converter. G1,G2and G3is the designation of each power converter, as in Fig. 10. adds virtual inertia capabilities, helping to reduce any frequency disturbancein the grid.Twofrequencyvariations(see Fig.15(a)) are induced by forcing some abrupt load changes (8.5 kW at 12 and−12 kW 18s)toobservethedynamicbehaviorof the control system. 1) Local Simulation: In Fig. 15, the response of the system is demonstrated, when a frequency variation is forced due to 6520 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 60, NO. 4, JULY/AUGUST 2024 Fig. 13. Simulation validation: (a) Active power reference tracking injected to the grid, taking into account the power-sharing between DEGs. (b) Reactive power reference tracking injected to the grid, taking into account the power-sharing between DEGs. G1,G2and G3is the designation of each power converter. DEG is the power developed by the power plant, as Fig. 10 shows. Ris the reference power. Fig. 14. Experimental validation: (a) Active power command, total power injectedto gridandpowerinjectedbyeachconverter.(b) Reactivepowerinjected to the grid, showing the command and the actual reactive power. The same legends as Fig. 13 are used in this plot. (c) and (d) Zoomed active and reactive power values and sent references. Actual is the power read by MODBUS TCP. Sent is the power sent by MODBUS TCP to the converter. a load-step change. Fig. 15(a) shows the frequency variation by using the SCAC system connected to the grid (showing the frequency with and without SCAC system). With SCAC system, the frequencies that appears are for two different rated power of the SCAC system. So frequency is forced to change at 12 s. In case of using the SCAC system, the ESS will inject active power (Fig. 13(a)) to the system in order to help the grid to increase its frequency, as it can be seen in b). On the other hand, if a suddenly frequency increase appears, as it can be seen at 18 s, the ESS-SCAC will absorb power from the system to decrease the frequency. Therefore, the power injection/absorption by the Fig. 15. Simulation: (a) Frequency variation of a weak grid due to a power change demand (with and without SCAC), modifying the installed power in the power plant. The higher power the lower frequency variation (f0is for the case without SCAC, f1is for the case with SCAC and Sn=6kVA,andf2is for the case with SCAC and Sn=10 kVA. (b) Power injected by each DEG, to mitigate the frequency variation from the case f1of b). G1,G2and G3are the designations of each power converter and DEG is the power of the total power plant, as Fig. 10 shows. SCAC system will depend on the inertia emulated and also the power installed in their ESS. It can be concluded from [13], the more power installed in the system, the lower the frequency variation will be. 2) Real-Time Experimental Test: Inthecaseofreal-timesimulation,theSCACsystemhasbeenemulatedbyusingthebattery module from CNG. Those results are presented in Fig. 16.Asit can be seen it works as the local simulation, when a frequency dip appears (Fig. 18(a)), DEGs inject power trying to reduce the frequency variation, respecting the power-sharing between converters (Fig. 18(b)).Inthis case, as a batteryemulator is used, the SOC state of each battery is presented in c), showing how the battery is charged or discharged. C. Case III This operation mode is controlled by the DEG operator (central controller), in order to reduce or increase the power injected by the power plant, depending on the grid requirements, as long as the ratings of the power plant are not exceeded. This means that if a power change is requested by the grid operator, the power injected will vary, taking into account that the RES are working normally at their maximum power point (MPP), and the excess or lack of power regarding the new power command will be managed by the SCAC-ESS. Once the grid operator’s setpoint returns to normal state, the storage system would stop absorbing energy, returning to zero power if there are not frequency changes. Another possible scenario is that the storage system reaches its maximum capacity and it cannot absorb more energy. This would mean that the PV string has to be taken out of its maximum power point to comply with the conditions of the grid operator. 1) Local Simulation: The aforementioned effect can be observed in Fig. 17, where at t =40 s, grid operator active power reference varies (PDEG, as shown in Fig. 15(a), forcing to inject less power from the DEG system. In this situation, either an