Maneuver decision is a critical factor which determines the success and failure of air combat for unmanned combat aerial vehicle. In order to increase the probability of wining air combat

a deep neural network (DNN) is proposed for maneuver decision. 36 kinds of maneuvers were constructed

and the samples of current situation

control quantity and future situation were acquired through flight simulations. The DNN is trained with the samples

making it capable of predicting future situation according to current information. Decision target function and situation assessment function were designed. In the process of air combat

the trained DNN is used to predict the future situations corresponding to all maneuvers

and the best maneuver is selected from all the maneuvers according to decision target function. The enemy planes which simply and autonomously maneuver were simulated

respectively

under different initial conditions

and the air combat situations were also assessed. The results show that the proposed decision method can be used to win air combat with less actions at balanced situation

and gain an edge through a series of actions at adverse situation

and the decision-making time is reduced by 9 ms.