Optimal bidding strategy for the price-maker virtual power plant in the day-ahead market based on multi-agent twin delayed deep deterministic policy gradient algorithm.

In recent years, distributed energy resources (DERs) have developed greatly, and the total capacity of resources aggregated by virtual power plants (VPPs) has increased. The role VPP plays in the market is changing. It is necessary to explore VPP's bidding strategy in the electricity market based on the impact of VPP on the clearing price. At the same time, the strategy game of different market players will also impact VPP's strategy. The multi-agent twin delayed deep deterministic policy gradient (MATD3) algorithm is used to solve the problem of price-maker VPP participation in the day-ahead (DA) market. VPP is required to submit the power and prices of electricity in multi-segments at different times. In the market bidding process, the main market participants are considered as agents. Iterative refinement of each player's strategy is achieved through multi-agent reinforcement learning, with the primary aim of maximizing revenue. For the VPP, the MATD3 algorithm improved the reward by approximately 65 % and increased the convergence speed by 17 % compared to the multi-agent deep deterministic policy gradient and multi-agent proximal policy optimization algorithms. The effects of different influences on VPP's bidding strategy are analyzed, which can provide a decision-making reference for market participants. • The VPP is seen as a price-maker when participating in the day-ahead market. • The TD3 algorithm is introduced for VPP to bid, enhancing bidding adaptability. • Multi-segment and multi-hour bidding strategies are examined. • The impact of various factors on the VPP's bidding strategies is analyzed. [ABSTRACT FROM AUTHOR]