Fit for purpose: modeling wholesale electricity markets realistically with multi-agent deep reinforcement learning

Abstract: Electricity markets need to continuously evolve to address the growing complexity of a predominantly renewable energy-driven, highly interconnected, and sector-integrated energy system. Simulation models allow testing market designs before implementation, which offers advantages for market robustness and efficiency. This work presents a novel approach to simulate the electricity market by using multi-agent deep reinforcement learning for representing revenue-maximizing market participants. The learning capability makes the agents highly adaptive, thereby facilitating a rigorous performance evaluation of market mechanisms under challenging yet practical conditions. Through distinct test cases that vary the number and size of learning agents in an energy-only market, we demonstrate the ability of the proposed method to diagnose market manipulation and reflect market liquidity. Our method is highly scalable, as demonstrated by a case study of the German wholesale energy market with 145 learning agents. This makes the model well-suited for analyzing large and complex electricity markets. The capability of the presented simulation approach facilitates market design analysis, thereby contributing to the establishment future-proof electricity markets to support the energy transition

Angaben zum Objekt

• Standort: Deutsche Nationalbibliothek Frankfurt am Main
• Umfang: Online-Ressource
• Sprache: Englisch
• Anmerkungen: ISSN: 2666-5468

Beteiligte, Orts- und Zeitangaben

• Ereignis: Veröffentlichung
• (wo): Freiburg
• (wer): Universität
• (wann): 2023
• Urheber: Harder, Nick ; Qussous, Ramiz Haidar Salim ; Weidlich, Anke
• Beteiligte Personen und Organisationen: Albert-Ludwigs-Universität Freiburg. Professur für Technologien der Energieverteilung

Weitere Informationen

• DOI: 10.1016/j.egyai.2023.100295
• URN: urn:nbn:de:bsz:25-freidok-2386771
• Rechteinformation: Open Access; Der Zugriff auf das Objekt ist unbeschränkt möglich.
• Letzte Aktualisierung: 14.08.2025, 10:52 MESZ