Offshore wind farms and hybrid wind-hydrogen plants derive revenue from multiple revenue sources, each subject to uncertainties and trade-offs. As a consequence, maximizing their overall profitability is challenging. Since electricity is typically traded ahead of its actual generation, weather forecasts play a crucial role in the power trading strategy. Additionally, the trading and control strategies of other market participants influence the balance of the public grid, affecting the revenue that can be generated by grid balancing. Moreover, the operational status of the electrolyzer may impact both the immediate and near-term hydrogen production potential.To address these challenges, we propose a novel multi-agent reinforcement learning (MARL) approach with two specialized reinforcement learning (RL) agents: one for day-ahead power trading, and a second for near-real-time power control of the wind farm and electrolyzer. The RL system is trained on SCADA data from a large offshore wind farm in the Belgian North Sea and on long-term power market data from the Belgian control area. Multiple scenarios with various electrolyzer ratings and hydrogen market prices are examined and compared.Results demonstrate the effectiveness of this approach, with the RL agents collaboratively maximizing the total operational profit of the hybrid plant, achieving a significantly higher profitability compared to conventional methods, both for the scenarios with and without hydrogen production. Furthermore, it is demonstrated that, despite the profit-maximizing objective for the RL agents, their policy does not affect overall grid stability significantly.
Ally, S, Verstraeten, T, Nowé, A & Helsen, J 2025, 'Day-ahead trading and power control for hybrid wind-hydrogen plants with multi-agent reinforcement learning', Applied Energy, vol. 401, no. part A, 126588, pp. 1-20. https://doi.org/10.1016/j.apenergy.2025.126588
Ally, S., Verstraeten, T., Nowé, A., & Helsen, J. (2025). Day-ahead trading and power control for hybrid wind-hydrogen plants with multi-agent reinforcement learning. Applied Energy, 401(part A), 1-20. Article 126588. https://doi.org/10.1016/j.apenergy.2025.126588
@article{5feb6640d2bb42a5984e30236b74c499,
title = "Day-ahead trading and power control for hybrid wind-hydrogen plants with multi-agent reinforcement learning",
abstract = "Offshore wind farms and hybrid wind-hydrogen plants derive revenue from multiple revenue sources, each subject to uncertainties and trade-offs. As a consequence, maximizing their overall profitability is challenging. Since electricity is typically traded ahead of its actual generation, weather forecasts play a crucial role in the power trading strategy. Additionally, the trading and control strategies of other market participants influence the balance of the public grid, affecting the revenue that can be generated by grid balancing. Moreover, the operational status of the electrolyzer may impact both the immediate and near-term hydrogen production potential.To address these challenges, we propose a novel multi-agent reinforcement learning (MARL) approach with two specialized reinforcement learning (RL) agents: one for day-ahead power trading, and a second for near-real-time power control of the wind farm and electrolyzer. The RL system is trained on SCADA data from a large offshore wind farm in the Belgian North Sea and on long-term power market data from the Belgian control area. Multiple scenarios with various electrolyzer ratings and hydrogen market prices are examined and compared.Results demonstrate the effectiveness of this approach, with the RL agents collaboratively maximizing the total operational profit of the hybrid plant, achieving a significantly higher profitability compared to conventional methods, both for the scenarios with and without hydrogen production. Furthermore, it is demonstrated that, despite the profit-maximizing objective for the RL agents, their policy does not affect overall grid stability significantly. ",
keywords = "Hybrid wind-hydrogen plant, Multi-agent Reinforcement Learning, Power trading, Power Control, Offshore wind energy",
author = "Stijn Ally and Timothy Verstraeten and Ann Now{\'e} and Jan Helsen",
note = "Publisher Copyright: {\textcopyright} 2025 The Authors",
year = "2025",
doi = "10.1016/j.apenergy.2025.126588",
language = "English",
volume = "401",
pages = "1--20",
journal = "Applied Energy",
issn = "0306-2619",
publisher = "Elsevier BV",
number = "part A",
}