Modeling Research On Missile Penetration And Combat Decision-making Based On Deep Reinforcement Learning

With the development of science and military,the advanced nature of weapons and equipment directly determines the status of the country in the world.The rapid change of weapons and equipment and the development trend of intelligent military have continuously expanded the scope of research and development.As a representative weapon for medium and long-range attacks,how to improve the intelligent level of missiles is also one of the hot spots in today’s weapons and equipment research.The content of this paper is aimed at the mobile penetration process during missile flight.On the platform of the combat simulation system of WESS,the red and blue sides are designed to fight according to the rules and to link operational theory.The main innovations are as follows:Firstly,carrying out intensive learning training based on combat simulation.For the different decision-making problems of intelligent penetration of missiles,training can be effectively carried out in accordance with different algorithm processes in experiments,corresponding interfaces are designed,the linkage of simulation platforms and algorithms is realized,and reinforcement learning algorithms are applied to simulation systems.Secondly,the simulation platform should integrate the model of the corresponding weapons and equipment to ensure the normal operation of the experiment and the validity of the data,and improve the credibility of the simulation platform.To realize the intelligent mobile penetration decision of warhead,using SAC algorithm to solve this experiment.Thirdly,applying multi-agent reinforcement learning algorithm to multi-target allocation.According to the centralized training method,designing the central processor model,and make sure about the relationship between agents,and the corresponding state input dimensions and action outputs are designed.The training results of the intelligent decision-making network are obtained,which proves the effectiveness of the multi-agent reinforcement learning using MASAC algorithm.