Research And Implementation Of Intelligent Decision-Making System For Wargame Based On Reinforcement Learning

With the rapid development of information technology,computer wargames have become an important means of simulating strategic decision-making,and artificial intelligence technology has made the fields of decision-making systems gradually move towards intelligence.Reinforcement learning,as one of the methods of machine learning,can autonomously learn the specified action strategy and complete the corresponding task through the interactive learning between the agent and the environment.Aiming at the problem of intelligent decision-making in tactical wargames,this paper studies from two aspects:adversarial decision-making and group games.Adversarial decision-making uses the known opponent’s behavior strategy to learn and guide the decisionmaking of its own deduction units independently,and obtain the optimal strategy.The group game is oriented to the scenario where multiple wargame players cooperate and confront each other,realize the autonomous game and strategy evolution,and provide strength comparison and game strategy analysis.This paper aims to study and implement an intelligent decisionmaking system for wargames based on reinforcement learning.Based on reinforcement learning technology,algorithms are designed to analyze the situation information processing of wargames,and effectively solve the problems of confrontation decision-making and group games in wargames.The key algorithms studied are as follows:For the adversarial decision-making problem,based on the Proximal policy optimization,an adversarial decision-making model combined with action mask and reward shaping is proposed,considering illegal actions and scene fog problems in wargames,to ensure the legitimacy of the strategy and enhance the exploration ability.Experiments show that performance of this model is better than that of several benchmark algorithms;for the group game problem,based on multi-agent reinforcement learning,a group game model is developed.Under the cooperative and confrontational relationship of multi-party wargames,the gamers are divided into different teams according to the game relationship,and the same reward scheme is adopted for the gamers of the same team.The group game algorithm sets an independent strategy for each player,combined with the twin value network and priority experience replay technology.The experiment shows that this model can play the game more effectively.This paper designs and implements an intelligent decision-making system for wargames based on reinforcement learning on key algorithms,providing human deduction and training and testing capabilities for intelligent confrontation decision-making and intelligent group games.