Intelligent multi-target cooperative assignment strategy based on deep reinforcement learning

An intelligent assignment strategy based on deep reinforcement learning is proposed for the multi-target allocation problem in interception. A comprehensive reward function is designed to improve the training efficiency under the concept of marginal reward in training. Meanwhile, to address the problem of poor searching ability induced by the huge solution space, the action masking module is introduced to modify the search strategy of the original deep Q-network (DQN) algorithm. The results show that the improved algorithm can achieve fast and smooth convergence. The trained strategy is generalizable and can generate a satisfactory allocation scheme in a short time.