A Wargame Agent Model Based on Multi-Agent Proximal Policy Optimization Algorithm

In recent years, wargame agents have attracted researchers' attention because of their potential application in the field of intelligent decision-making. The construction of wargame agent faces many challenges, such as the complexity of environmental information, the high dimension of decision space and the difficulty of multi-agent cooperation. The development of multi-Agent reinforcement learning (MARL) technology provides an effective way to deal with these challenges. Based on the multiagent proximal policy optimization algorithm (MAPPO), this paper designs a multi-agent model of wargame. We redesign the dynamic reward function to balance short-term benefits and longterm goals, and adopt a two-stage training strategy to improve the training efficiency of the model. Finally, the effectiveness of the proposed model is verified by experiments on the wargame deduction system.