A multi-agent game model for agent behavior analysis was proposed to solve the problems of strong coupling, fast strategy feedback and obvious price fluctuation among power generators, electricity selling companies, load aggregators and large users in the electricity market. In this method, the market state coding, agent payoff modeling, optimal response mechanism and iterative solution process are integrated into a unified computing framework to describe the evolution law of multi-agent strategies in continuous trading environment. The simulation results based on 120 trading days and 56 market players show that the social welfare of the model reaches 91.8, the market efficiency reaches 0.943, enters the stable range after an average of 735 rounds, and the price volatility coefficient decreases to 0.118. The overall performance of the model is better than that of the single-agent reinforcement learning, non-cooperative game and centralized optimization model. The research shows that the method can better reveal the formation mechanism of power market main body behavior, and provide computational support for market operation analysis and intelligent decision-making.
