基于复合回报函数的空战指向控制策略研究OA

In allusion to the problem that the rapid pointing control of unmanned combat air vehicle(UCAVs)in close air combat is difficult to be in any posture,an air combat pointing control strategy based on the design of composite payoff function is proposed.In order to avoid the dangerous state of energy retreat after the UCAV completes the pointing task with autonomous inefficient large maneuver in air combat,a composite payoff function based on multi-dimensional constraints such as quantitative energy,time,and angle of attack is designed for pointing and aiming training of UCAV in different initial postures.In allusion to the problem of difficult convergence of strategies caused by the complex high-dimensionality of observation space and action space in air combat missions,the network parameter updating process of the dual Actor-Critic neural network structure in the training of SAC algorithm is subjected to hierarchical L2-paradigm gradient cropping,which substantially improves the convergence efficiency of the algorithm.The simulation results show that the proposed algorithm can well guide the aircraft to quickly make maneuvering decision commands to preserve energy and maneuverability to complete the pointing and aiming task,and has better convergence efficiency than traditional deep reinforcement learning algorithms,such as TD3,PPO,DDPG,and so on.