基于深度强化学习的高超声速飞行器动态面控制方法

Dynamic Surface Control Method for Hypersonic Vehicle Based on Deep Reinforcement Learning

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,基于模型先验知识,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所设计控制方法的有效性与鲁棒性。

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.