Virtual power plant market trading strategy based on hybrid game reinforcement learning

With the rapid development of regional distributed energy, the issues of small installed capacity and strong output variability have become increasingly prominent, resulting in insufficient competitiveness when distributed energy participates in market transactions independently. To enhance its market participation capabilities, integrating distributed energy resources into virtual power plant has emerged as an effective approach. Therefore, this study investigates market trading strategies for virtual power plant incorporating distributed energy resources and proposes a trading strategy based on hybrid game-based reinforcement learning. First, establish revenue models for energy suppliers and load aggregators based on the operational characteristics of internal units within the virtual power plant. Then, to ensure the overall profitability of operators within the virtual power plant, a social welfare maximization model is established. Finally, the transaction model is solved using a hybrid game-based reinforcement learning algorithm combining Stackelberg and evolutionary game theory. Case studies demonstrate that the two-layer model based on hybrid game-theoretic reinforcement learning algorithms outperforms traditional intelligent algorithms, reducing computation time by nearly 50%. Furthermore, when virtual power plants participate in both energy markets and ancillary service markets, they can achieve higher returns.