Research on Intelligent Operational Assisted Decision-making of Naval Battlefield Based on Deep Reinforcement Learning

∗With the increasingly complex naval battlefields and the rapid development of artificial intelligence in the future, it has become an inevitable trend for the naval battlefield combat aid decision-making to develop toward intelligence. The purpose of the research is to embed the simulation test environment based on deep reinforcement learning technology into the combat auxiliary decision-making system, use simulation to support ongoing military decision-making operations, and provide an auxiliary decision-making reference for the commander’s multi-branch plan real-time decision-making in the emergency combat environment. Combining deep reinforcement learning and Monte Carlo tree search, the strategy network selects decision branches to reduce the search width, and the value network evaluates the naval battlefield situation to reduce the search depth. Meanwile, the self-game of reinforcement learning is used to adjust the strategy network, improve the performance of the strategy network, and use adversarial deduction to further train the value network. Finally, when the next branch decision is made, the intelligent simulation engine is used to determine the optimal branch decision under the current situation by combining the Monte Carlo tree search algorithm of the strategy network and the value network. The complexity of the information-based naval battlefield determines the importance of improving the ability to assist in combat decision-making. Research and explore the use of artificial intelligence as a commander’s assistant for real-time combat assistance decision-making, and make a way to solve the difficulties and challenges of intelligent decision-making in naval battlefields.