Multi-constrained intelligent gliding guidance via optimal control and DQN  

作　　者：Jianwen ZHU Hao ZHANG Sibo ZHAO Weimin BAO 

机构地区：[1]School of Aerospace Science and Technology,Xidian University,Xi’an 710126,China [2]China Aerospace Science and Technology Corporation,Beijing 100048,China

出　　处：《Science China(Information Sciences)》2023年第3期210-225,共16页中国科学（信息科学）（英文版）

摘　　要：In order to improve the adaptability and robustness of gliding guidance under complex environments and multiple constraints,this study proposes an intelligent gliding guidance strategy based on the optimal guidance,predictor-corrector technique,and deep reinforcement learning(DRL).Longitudinal optimal guidance was introduced to satisfy the altitude and velocity inclination constraints,and lateral maneuvering was used to control the terminal velocity magnitude and position.The maneuvering amplitude was calculated by the analytical prediction of the terminal velocity,and the direction was learned and determined by the deep Q-learning network(DQN).In the direction decision model construction,the state and action spaces were designed based on the flight status and maneuvering direction,and a reward function was proposed using the terminal predicted state and terminal constraints.For DQN training,initial data samples were generated based on the heading-error corridor,and the experience replay pool was managed according to the terminal guidance error.The simulation results show that the intelligent gliding guidance strategy can satisfy various terminal constraints with high precision and ensure adaptability and robustness under large deviations.

关 键 词：gliding flight optimal guidance velocity control deep reinforcement learning intelligent decision 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]