IEA Wind TCP Task 51 "Forecasting for the weather driven energy system" - 2nd Austrian Workshop

Aktivitäten und Highlights aus dem internationalen IEA Task 51 Fokus Extreme Meteorological Events for Power Systems Workshop 11.11.2025 Irene Schicker [email protected] Was hat sich getan im internationalen Task 2026 © GeoSphere Austria 2 Jänner 2025 Online Task Meeting mit Beiträgen: •Lukas Strauss: Considerations on AI safety for future integrated forecasting systems. •Irene Schicker: Summary of the 1st Austrian IEA Task 51 workshop –methodologies applied, lessons learnt. 11. Jänner 2025 Webinar Deep Learning Weather Prediction: Nachschau ist hier möglich: https://www.youtube.com/watch?v=t6H7diavQdg&t=2825s&pp=0gcJCQMKAYcqIYzv Oktober 2025 Nachholen des von April in Zeit und Ort verschobenen Workshops „ Forecasting Extremes in the power system “. Nachschau hier am Youtube Kanal möglich: https://www.youtube.com/c/IEAWindForecasting November 2025 • Webinar 1: 5th November 2025 - Advances in WeatherAI • Webinar 2: 12th November 2025 - Application of Meteorological Technology in Electric Power Systems: new IEC Sub-Committee • Webinar 3: 19th November 2025 - Verification, Validation & Uncertainty Quantification • Webinar 4: 26th November 2025 - Extreme Events in the Power System: Workshop Results & next steps https://dtudk.zoom.us/j/62313644719?pwd=gkyuFSzaD4EawvOqHqDH9UgRomkLaW.1&from =addon Workshop Overview and Context – Why it matters now © GeoSphere Austria 3 Das Setting: •60+ international experts from met services (private, governmental), TSOs, energy companies, research institutions, traders •Core challenge: High renewable penetration = high weather sensitivity •New paradigm: "Extreme" = system stress, not just meteorological rarity •Three critical aspects: threatening infrastructure, challenging supply security, forecast errors and their impact on trading, decision making, grid security Defining „Extremes“ – wie kann man sie definieren? © GeoSphere Austria 4 Multi-Dimensional Event Classification Simple events: Single parameter extremes (e.g. high wind >25 m/s) Compound events: Multiple simultaneous extremes (e.g. Dunkelflaute) Consecutive events: Sequential extremes amplifying impacts (e.g. heat) Cascading events: Weather triggers causing system-wide failures Climatological/Meteorological Extreme Events Derived from climate data or (forecast) model simulations, or observations Defined relative to climatology (30-year period) or thresholds (e.g. survival wind speed) Probabilistic in nature (characterized by return periods) May exhibit long-term changes (duration, intensity, location shifts) due to climate change Characterized for one or combined variables or caused by a variety of extreme events Acute event vs chronic hazards vs returning | Extremes – Variability – Volatility Can be location/region specific but impacts can be on a larger scale But… there are events that can also be high impact/adverse weather and not related to meteorological extremes Climatological/Meteorological Extreme EventsHigh Impact Weather (HIW) Events Power System Extremes Atmospheric extremes can provide initial insight, may overlap with or differ from extreme operating conditions. Need additional information from asset managers, engineers, system operators to understand what may be considered “extreme operating conditions for power system assets or components”. Define performance thresholds relative to the operating/infrastructure characteristics of assets/system Can be locationand asset specific What is considered extreme may change through the adoption of new technology or operating methods (and climate change), continuously update definitions Can be caused by internal equipment failure, malfunctions, supply shortages, human error, manmade attacks… Goal: ensure that power system assets can maintain adequate operations or ensure rapid recovery Meteorological vs Power System Extremes - Examples Changes in extreme events are likely to be more impactful than changes in average conditions Temperature Extremes Precipitation Extremes Wind Extremes historical 95th/99th percentile of summer daily high 100 - year rainfall / 13 mm in a day Storm systems, fast running low pressure systems, thunderstorms, derechos, gusts Distribution transformers are at risk of failure if subjected to 2 -3 consecutive days of > 40 C Transmission line management(?) Sensitive electronics can fail if submerged in floods. Severity of flood impact depends on the elevation/waterproofing of equipment . Foundation of infrastructure might be eroded. Extreme wind gusts damaging power lines, which in turn can cause wildfires; turbines unable to operate > 25 m/s if they don’t have storm control ; wind load on infrastructure, trees,… Climatological extremes Power system extremes Critial Event Types – From Dunkelflaute to Hellsturm to Brownout to... © GeoSphere Austria 8 Dunkelflaute Combined capacity factor <0.06 for ≥48h, median 3.2 days (max 8 days) Wind Ramping Minor (3-5 m/s) to Extreme (≥10 m/s) over 3 hours, ±500 MW fluctuations Wind Drought Persistent low winds (<3 m/s), 7-32 days duration Hellsturm High wind + high solar + low demand, curtailment challenges Event Type Duration Key Metric Dunkelflaute 2 -8 days Capacity <0.06 Wind Ramping 3 hours ± 500 MW Wind Drought 7 -32 days <3 m/s Hellsturm Hours Curtailment Wind turbines (IEC 61400-1): Cut-in 3 m/s, rated 12-15 m/s, cut-out 25 m/s, structural risk >60 m/s PV systems (IEC 61215): Performance losses >30°C, operational range -40°C to +85°C Grid infrastructure: Icing, galloping power lines, flooding thresholds Innovation: Direct mapping meteorological → operational risk Parameter Threshold Standard Wind Cut -in 3 m/s IEC 61400 -1 Wind Cut - out 25 m/s IEC 61400 -1 Wind Structural >60 m/s IEC 61400 -1 PV Performance Loss >30 °C IEC 61215 PV Operating Range - 40 to +85°C IEC 61215 Infrastructure calibrated thresholds – Linking Meteorology to Engineering Standards EPRI READi Framework for Long-Term Resilience Three-pillar framework: Physical climate data → Asset vulnerability → System modeling & investment Comprehensive resources: Climate 101 training (1000+ participants), interactive story maps, climate chatbot Multi-year events: Workshop consensus that 1several year energy shortages should inform planning Freely accessible: epri.com/readi Pillar Description Physical Climate Data Climate data inventory, user guides, gap assessments Asset Vulnerability Hazard-exposurevulnerability assessments System Modeling Climateinformed power system modeling, prioritization From forecast to the climate scales – how can we ensure long-term resilience Secure AI & Robustheit in der Energieprognose Systemisches Risiko: Forecast Harmonization Wenn alle TSOs, Trader, Dispatchers dasselbe AI-Modell nutzen → synchronized forecast errors → Blackout-Risiko AI Limitations Tail Risk Underestimation AI-Modelle trainiert auf ERA5 → Extrema (>99th pctl) unterrepräsentiert Black Box Problem Keine physikalische Erklärbarkeit → Vertrauensproblem bei TSOs Data Dependency Out-of-distribution: Klimawandel → neue Extrema außerhalb Training-Data Multi-Provider Strategy Governmental Services DWD (ICON), ECMWF (IFS), MeteoSwiss, GeoSphere (AROME) Private Anbieter Meteomatics, weather.com, Energy & Meteo Systems (EMS) AI-Modelle (experimentell) GraphCast, Pangu als Ergänzung, NICHT als Ersatz Best Practices Hybrid Forecasting NWP (physics-based) + AI (pattern) + Statistical PP → Consensus Forecast Ensemble Diversity Nie nur 1 Modell für kritische Entscheidungen! Min. 3 unabhängige Quellen Explainable AI (XAI) SHAP, LIME für Feature-Attribution → TSOVertrauen Workshop Consensus: "Never rely on single forecast source for grid operations" - Critical infrastructure requires redundancy! Five Essential Messages for Austrian Energy Transition 1. "Extreme" = system stress: Include compound events and forecast errors, not just storms 2. Multi-scale capability: We can act across all time scales (minutes → seasons → climate) 3. Standardization matters: IEA Task 51 framework enables international benchmarking in connection with the IEA Subcommitte on High Impact Weather 4. Forecast diversity is critical: Avoid harmonization risks through multiple providers 5. Austria's unique position: Alpine expertise + multi-scale approach Key Takeaways Workshop Offenbach Webinar AI, Machine Learning in der Extremereignis-Prognose AI Weather Prediction (AIWP) GraphCast (Google DeepMind), Pangu-Weather (Huawei), FourCastNet (NVIDIA) - neue Ära der NWP Vorteile: 1000x schneller, ähnliche Skill wie IFS (bis Tag 7) Challenge: Extrema-Underrepresentation, physikalische Konsistenz Webinar 1: 5th November 2025 - Advances in WeatherAI Kritische Erkenntnis: AI-Modelle sind schnell & gut für "normale" Bedingungen, aber können Extrema überoder unterschätzen. Hybrid-Ansatz (NWP + AI) ist für kritische Infrastruktur essentiell! https://www.youtube.com/watch?v=qB4japkodCs © GeoSphere Austria 20 Was ist für EUER Energiesystem die größte Herausforderung bei Wetterextremen? The Slido app must be installed on every computer you’re presenting from © GeoSphere Austria 21 In einem Wort: Was macht alpine Wetterextreme besonders herausfordernd? The Slido app must be installed on every computer you’re presenting from © GeoSphere Austria 22 Wie viele UNABHÄNGIGE WetterprognoseQuellen nutzt eure Organisation für kritische Entscheidungen? The Slido app must be installed on every computer you’re presenting from © GeoSphere Austria 23 Was wäre euer WUNSCH-Feature für ein perfektes meteorologisch angetriebenes Extremereignis im System? Von früher Detektion, zeitliche und räumliche Auflösung... The Slido app must be installed on every computer you’re presenting from © GeoSphere Austria 24 Audience Q&A was removed The Slido app must be installed on every computer you’re presenting from