关键词: 定向能系统, 反无人机集群, 深度Q网络, 联合机制, 威胁评估机制, 动作屏蔽机制

Abstract:

In order to solve the problem of dynamically addressing firepower intelligent decision-making in anti-UAV cluster combat using a directed energy system, a deep reinforcement learning model is established. Based on the high multi-agent state and action space dimensions of this model, a modeling and simulation method of firepower intelligent decision-making of directed energy system based on joint deep Q network (DQN) is proposed. The state space is constructed from the state of directed energy system, UAV cluster and the directed energy system deployment area. The joint mechanism is used to share the state information of each equipment and the network parameters of the same type of equipment. The threat assessment mechanism is designed to improve generalization, and the action shielding mechanism is established to shield invalid actions. The problems of divergence and slow convergence of multi-agent training, caused by state and action dimension disasters, are effectively solved, and the learning efficiency and generalization of Joint DQN network are improved. According to the simulation results, this method is superior to the traditional rule-based method, which verifies the feasibility and practicability of this method, and provides a new idea for intelligent decision-making of anti-UAV cluster firepower of directed energy system compatible with various deployment schemes.

Key words: directed energy system, anti-UAV cluster, deep Q network(DQN), joint mechanism, threat assessment mechanism, action shielding mechanism