Abstract:

In order to solve the problem of complex combat information and difficult to quickly and accurately perceive situation and make decision in air combat, an algorithm combining game theory and deep reinforcement learning is proposed. Firstly, according to the typical one-to-one air combat process and the standard of random game, a two machine multi-state game model under the condition of red and blue confrontation in close air combat is constructed. Secondly, deep Q network (DQN) is used to deal with the continuous infinite state space of fighter. Then, the Minimax algorithm is used to construct a linear programming to solve the optimal value function of the stage game in each specific state, and the network approximation value function is trained. Finally, the optimal maneuver strategy is obtained according to the output of the network after training. The simulation results show that the algorithm has good adaptability and intelligence for the air combat. It can effectively select the favorable maneuver action and occupy the dominant position according to the air combat opponent's action strategy.

Key words: game theory, deep reinforcement learning, stochastic game, air combat strategy