关键词: 高超声速飞行器, 制导控制一体化, 规避与突防, 深度强化学习

Abstract: Aiming at the problem of no-fly zone and interceptor avoidance during reentry phase, a learning-based integrated fault-tolerant guidance and control method for hypersonic vehicles considering avoidance and penetration is proposed. The proposed method is based on the idea of separate longitudinal and lateral guidance design in traditional predic-tive correction guidance methods. First, in longitudinal guidance, to improve the guidance performance of the vehicle, based on the integrated design concept of guidance and control, the six degree of freedom model of the vehicle, in-cluding the attitude fault-tolerant control system, is adopted for predicting the range-to-go. In view of the resulting cal-culation time-consuming problem, the deep network is further used to fit the predicting process to improve the real-time performance of the algorithm. Then, in lateral guidance, to solve the problem of avoiding no-fly zone and inter-ceptor, the interfered fluid dynamical system is introduced to calculate the expected heading angle of the vehicle con-sidering the threat factors, and the heading angle error corridor and the flip logic of the bank angle are combined to obtain the command, so that the vehicle has the ability of avoidance and penetration. Then, considering the complexi-ty of flight environment, the difference of maneuvering ability of vehicle during reentry phase, and the influence of dy-namic characteristics of attitude control system on guidance performance, the deep reinforcement learning algorithm is combined to conduct agent training, so that it can make online decisions on key parameters of algorithm according to the real-time state of vehicle, and improve the effectiveness and adaptability of the proposed guidance method. Finally, the results of semi-physical comparison simulation and Monte Carlo simulation show that the proposed meth-od can achieve strong avoidance, penetration and guidance performance.

Key words: Hypersonic vehicle, Integrated guidance and control, Avoidance and penetration, Deep reinforcement learning