Research on Online Firepower Distribution of Multi-bomb Cooperative Anti-ship Based on MADRL

In modern naval warfare, the attack and defense between anti-ship missile and target ship has always been an important problem. Whether the local ship formation can be attacked quickly and effectively has become the winning point of leading naval warfare. Anti ship missile with its high precision, high range, high damage rate has become one of the main weapons in today’s maritime operations. However, due to the limited damage ability of single missile, multi missile cooperative attack mode is often adopted in actual combat. Therefore, how to reasonably distribute the firepower of anti-ship missile is often the key problem in modern naval warfare. In this paper, a new method based on deep Q-value network is proposed to solve the fire allocation strategy of anti-ship missile. Unlike the traditional algorithms in this field, DQN algorithm can automatically solve the optimal fire allocation strategy by interacting with the environment without relying on any prior information of the enemy. The algorithm fully gives full play to the computational ability of neural network and the ability of decision-making of reinforcement learning, solves the problem that the traditional algorithm can not realize rapid decision-making, and also solves the problem that the traditional reinforcement learning algorithm cannot traverse because of the large state space. The experimental results show that the deep Q-value network algorithm gives the best fire allocation scheme with the best damage effect under different formation and training times.