| 基于深度强化学习的高超声速飞行器动态面控制方法 |
| |
| 作者姓名： | 胥 彪 赵琛钰 李 爽 朱东方 |
| |
| 作者单位： | 南京航空航天大学 航天学院；
南京航空航天大学空间光电探测与感知工业和信息化部重点实验室;上海航天控制技术研究所 |
| |
| 基金项目： | 国家自然科学基金（61603183）；南京航空航天大学空间光电探测与感知工业和信息化部重点实验室开放课题资助（NJ2022025-6）；中央高校基本科研业务费资助（NJ2022025） |
| |
| 摘 要： | 针对高超声速飞行器控制问题，通过将深度强化学习与动态面控制方法相结合，设计了智能姿态控制算法。首先，利用模型先验知识，采用传统的动态面控制方法设计控制器结构。然后，考虑跟踪误差和控制量幅值约束的指标情况下，采用深度学习算法完成对控制器参数的智能寻优，代替传统设计中的人工调参试错过程。为提升训练效果，在奖励函数中引入了控制量变化率。最后，通过数值仿真验证了本文所设计控制方法的有效性与鲁棒性。
|
| 关 键 词： | 高超声速飞行器 动态面控制 深度强化学习 智能优化 |
Dynamic Surface Control Method for Hypersonic Vehicle Based on Deep Reinforcement Learning |
| |
| Authors: | XU Biao ZHAO Chenyu LI Shuang ZHU Dongfang |
| |
| Institution: | College of Astronautics, Nanjing University of Aeronautics and Astronautics;
Key Laboratory of Space Photoelectric Detection and Perception (Nanjing University of Aeronautics and Astronautics),
Ministry of Industry and Information Technology; Shanghai Aerospace Control Technology Institute |
| |
| Abstract: | For the control problem of hypersonic vehicle, an intelligent attitude control algorithm is designed by deep reinforcement learning and dynamic surface control method. First, based on the model prior knowledge, the control structure is designed by dynamic surface control method. Then, by considering the tracking error and the constraint of the control magnitude, the deep learning algorithm is used to complete the intelligent optimization on the controller parameters, instead of the manual tuning in the traditional design procedure. In order to improve the training effect, the change rate of control variable is introduced into the reward function. Finally, the effectiveness and robustness of the proposed control method are verified by numerical simulation. |
| |
| Keywords: | hypersonic vehicle dynamic surface control deep reinforcement learning intelligent optimization |
|
| 点击此处可从《》浏览原始摘要信息 |
| 点击此处可从《》下载免费的PDF全文 |
