Strategic bidding of wind farms in medium-to-long-term rolling transactions: A bi-level multi-agent deep reinforcement learning approach

The increasing penetration of renewable energy in the electricity market suppresses marginal prices, posing profitability challenges for wind power producers. To address this, effective medium-to-long-term (MLT) rolling transactions can hedge against spot market price risks and improve profitability. However, conventional bidding approaches often fail to capture the intricate uncertainties associated with wind generation and trading dynamics over extended periods. This paper introduces a bi-level multi-agent deep reinforcement learning (DRL) approach specifically designed for optimizing wind energy MLT rolling transactions. The proposed method innovatively integrates the Black–Scholes model with the Hamiltonian function to structure an optimal decision-making framework that balances short-term bidding efficiency with long-term strategic positioning. By separately optimizing transaction quantities and prices, the model prevents conflicts between these variables and ensures more accurate and effective decision-making. Additionally, the approach leverages advanced spatiotemporal modeling capabilities through the TimesNet-Latent-GNN framework, enabling it to capture complex market dependencies and achieve superior performance in managing price risks and maximizing profitability. Validation using real-world transaction data from the Shanxi electricity market demonstrates that the proposed method significantly outperforms traditional risk-averse strategies in terms of profitability and risk mitigation.