Agent Optimization Decision-Making Method Based on Self-Play

Aiming at the problem that the traditional deep reinforcement learning algorithm has poor generalization ability of generating strategies when solving complex game problems such as real-time strategy games, an agent strategy optimization method based on the combination of expert knowledge and self-play is proposed. Firstly, the opponent strategy of the agent is designed based on expert knowledge, and the corresponding strategy is generated by the proximal policy optimization algorithm (PPO), and the generated strategy is continuously put into the opponent strategy pool to provide a variety of opponent strategies for the agent in the training process. Through continuous iterative confrontation, the agent's autonomous decision-making ability in the face of different opponent strategies is improved. The simulation results show that compared with the traditional deep reinforcement learning algorithm, the introduction of the self-play training framework can significantly improve the average confrontation rate of the agent strategy, improve the robustness and generalization ability of the agent confrontation strategy, guide the agent to learn more effective tactical strategy, and achieve the adversarial advantage of “Priority first enemy discovery, Priority enemy launch”.