Deep Reinforcement Learning-Based Radar Network Target Assignment

This study focuses on the problem of target assignment when a phased-array radar network detects hypersonic-glide vehicles in near-space and proposes a method for target assignment based on deep reinforcement learning. The state, action, and reward functions of the agent and the structure of the deep Q network are designed. To solve the problem of the large scale of the agent’s action space, an incremental search method is proposed. The proposed incremental search method reduces the scale of the search space by at least 16 orders of magnitude. To improve the agent’s solution quality during a search, an agent soft search constraint is designed. Setting the soft search constraint filters out the obviously inferior solutions, thereby improving the solution quality. The performance of the proposed method is verified in two scenarios: a multitarget attack and a saturation attack. The simulation results showed that the performance of the method proposed in this paper is significantly better than those achieved by the heuristic algorithm and the random assignment method in terms of the number of target assignments, the target threat degree, the number of radar switches and the cumulative detection duration.