Optimization Framework for Fire Allocation Against Large-Scale Naval Vessels Using Deep Q-Networks

Traditional fire allocation struggles to model physical coupling effects in large-scale naval vessels. This paper proposes a reinforcement learning method using damage tree analysis to overcome this limitation. It formulates the fire allocation problem as a cost-minimization optimization problem subject to damage threshold constraints. Furthermore, the Deep Q-Network (DQN) algorithm is introduced, incorporating an Action Masking mechanism to constrain the agent's decision-making within a valid action space. Simulation results demonstrate that this method exhibits significant advantages in optimizing the quality of fire allocation plans, reducing ammunition consumption, and improving computational efficiency. Compared to traditional heuristic algorithms (such as Ant Colony Optimization and Simulated Annealing), the DQN algorithm can rapidly generate efficient and economical strike plans.