Assessing flexibility option potential by combining electricity price forecasting and agent-based electricity market modelling

Kurzfassung

Motivation Future energy systems with high shares of variable renewable energy sources will increasingly rely on flexibility options (FO) such as battery storage, pumped-hydro plants, and demand response [1]. The competitive dynamics between these technologies will shape electricity market outcomes and price formation, making it essential to understand their operation and market behavior [2]. We present a novel modeling workflow that combines machine learning with agent-based modeling to address this challenge. Our approach provides real-time electricity price forecasts to a fundamental electricity market model, enabling comprehensive assessment of different flexibility options and their market impacts. Method We apply the open[1] agent-based electricity market model AMIRIS [3]. This model captures the heterogeneous characteristics of market participants, including their objectives, risk preferences, information asymmetries, and strategic interactions within the market environment. To enhance the model's representation of competing FOs, we introduce AMIRIS-PriceForecast [4], a dedicated electricity price forecasting module that integrates multiple machine learning architectures in a modular way, supplying time series forecasts directly to AMIRIS. Unlike conventional approaches that rely on historical market data, we train a Temporal Fusion Transformer [5] on future electricity market scenarios, explicitly accounting for evolving market dynamics driven by the ongoing energy transition [6]. Communication across models is facilitated through FastAPI[2], ensuring minimal computational overhead. Results The FO applies the machine learning-based electricity price forecast to optimize its own dispatch operation. We analyze the storage activity of a risk-taking FO with a 1 MW/5 MWh electricity storage as a price-taker in the day-ahead market. The results demonstrate a characteristic charging and discharging pattern: the storage operator charges during nighttime and midday periods when electricity prices are typically lower, and discharges during morning and evening peak hours when prices are higher. However, when compared to a perfect foresight benchmark, the analysis reveals some situations with suboptimal operational decisions resulting from forecasting inaccuracies. These forecasting errors significantly reduce the overall profitability. Our approach also enables the simulation of competing FOs. In scenarios with homogeneous FO concentration, individual operator revenues decline once market saturation is reached. This occurs due to diminishing price spreads caused by the collective impact of FO operations on electricity prices. When heterogeneous FOs are active on the market, the dynamics shift: certain operator, i.e. particularly high-powered short-duration systems, first benefit from increased competition. Nevertheless, once available FO capacity reaches saturation, overall market profits decline sharply. Future research should a) examine additional revenue streams from, e.g., intraday markets, b) enhance operational strategies that enable FOs to account for their market price impacts, and c) account for latest advancements in timeseries forecasting. 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).   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/s40518-018-0092-x. [2]   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. [3]   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. [4]   F. Nitsch and C. Schimeczek, "AMIRIS-PriceForecast," 2025, doi: 10.5281/zenodo.14907870. [5]   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. [6]   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.   [1] gitlab.com/dlr-ve/esy/amiris (accessed on 23rd May 2025) [2] github.com/fastapi/fastapi (accessed on 23rd May 2025)