Simulation of Evolutionary Game Decision Model Based on Reinforcement Learning Algorithm

Evolutionary game theory is different from traditional game theory. The key of evolutionary game analysis is to determine the model or mechanism of agent learning and strategy adjustment. RL (Reinforcement Learning) is a widely used machine learning algorithm, which can deal with uncertain information in dynamic environment and make agents choose the best strategy. RL algorithms can be divided into model-free RL algorithm and model-based RL algorithm. Because RL can get more information through trial and error and dynamic environment interaction, the selected action can achieve the maximum cumulative return of the environment. In this paper, Q-Learning algorithm and game model are combined to form a dynamic evolutionary game decision-making model with learning ability, which can break through the limitation of Q-Learning algorithm single agent. The simulation results show that by introducing Q-Learning, this model can distinguish enterprises with good reputation decision-making style from enterprises with bad reputation decision-making style through income, which has a good distinction and simulates real economic activities well. The algorithm of optimal decision-making by constantly learning environmental knowledge enriches the content of game decision-making problems to some extent.