A Weapon Target Assignment Algorithm Based on Dueling DQN

In response to the escalating challenge posed by hypersonic weapons, characterized by their high velocity and unpredictable trajectories, there is a pressing need to develop fast and efficient Weapon Target Assignment (WTA) algorithms. The precise algorithms are difficult to deal with large-scale problems and the heuristic algorithms demand substantial computational resources. Therefore, this paper proposes a WTA algorithm based on improved Dueling DQN. We first build a WTA model based on the maximum damage probability criterion and convert it into a multi-stage decision process. Then we design an improved Dueling DQN algorithm framework. To ensure the independence of the Dueling DQN network structure from the number of targets or interceptors, we design the state vector and action vector, where the state vector is equivalent to the state, and the action vector corresponds to the action one by one. The state vector serves as the input for the value network, whereas the state vector and action vector, in tandem, are the inputs for the advantage network. This design ensures that the information pertaining to each target is decoupled from the interceptors. The experimental findings substantiate the superior performance of the proposed algorithm. When tackling large-scale WTA problems, our method not only matches but often outperforms heuristic algorithms in terms of both solution quality and decision-making speed. Particularly noteworthy is the significant advantage in computational speed over heuristic methods such as genetic algorithms.