Optimal Guidance for Orbital Pursuit-Evasion Games Based on Deep Neural Network

Integrating the artificial intelligence into space missions is attracting increasing attention from scholars. This paper concerns on the optimal guidance problem of orbital pursuit-evasion games, and an optimization method based on the deep neural network (DNN) is proposed to improve the efficiency of solution. First, the problem is formulated by a zero-sum differential game model, which transforms the original problem to a TPBVP. Second, we propose an optimization method using a DNN to generate individual guesses for further optimization through a gradient-based local optimization algorithm. Finally, numerical simulation results show that, after training the DNN with samples generated through the traditional method, the proposed optimization method statistically improves the efficiency over the traditional optimization by roughly two orders of magnitude without losing quality, and it is feasible in different cases.