面 向 多 战 术 需 求 的 无 人 机 空 战 自 主 规 避 机 动 方 法

Air combat is usually a continuous process involving multiple rounds of missile confrontation. Unmanned Combat Aerial Vehicle（UCAV）should comprehensively consider the impact of maneuvering on the entire air combat mission in the process of evading incoming air-to-air missiles，instead of focusing only on safety factors. In this paper，a UCAV autonomous evasive maneuver method is proposed under the condition of multi-tactical requirements such as miss distance，energy consumption and terminal superiority. A three-dimensional pursuit and escape model of UCAV-missile and a model for the state space，action space and reward function of UCAV autonomous evasion are established. An algorithm based on the LSTM-Dueling DDQN（Long Short-Term Memory-Dueling Double Deep-Q Network）is proposed for this model. The algorithm fuses Double DQN and Dueling DQN network models，and uses LSTM network to extract timing features. Based on the concept of exploratory course learning，temporal fusion of dense and sparse reward functions is carried out to promote joint guidance of artificial experience and strategy exploration in the process of maneuver learning. The Chebyshev method is introduced to solve the Pareto solution set for different degree of tactical demands，so as to reflect the contradiction and coupling of multiple tactical requirements. Simulation experiments and result analysis show that the proposed method has good convergence speed and learning effect，and is feasible and effective to solve the problem of autonomous evasive maneuver in air combat under multiple tactical requirements. The obtained evasive maneuvers can reflect different evasive tactical requirements while ensuring UCAV’s own safety.