引入积分补偿的四旋翼确定性策略梯度控制器  

Deterministic policy gradient controller with integral compensator for quadrotor

作　　者：孙丹 高东 郑建华 韩鹏 SUN Dan;GAO Dong;ZHENG Jian-hua;HAN Peng(National Space Science Center,Chinese Academy of Sciences,Beijing 100190,China;National Space Science Center,University of Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]中国科学院国家空间科学中心,北京100190 [2]中国科学院大学国家空间科学中心,北京100190

出　　处：《计算机工程与设计》2023年第1期255-261,共7页Computer Engineering and Design

基　　金：北京市科技计划基金项目(Z191100004319004)。

摘　　要：为实现四旋翼无人机位置姿态的自主智能控制,结合强化学习中深度确定性策略梯度算法设计无人机智能控制器,对该控制器在四旋翼位置跟踪过程中出现静差的现象进行分析,提出带积分补偿的改进深度确定性策略梯度算法。仿真结果表明,引入积分补偿的改进深度确定性策略梯度控制器能够实现四旋翼位置姿态的自主稳定控制,位置误差为零,在奖励函数变化后控制过程依然保持平稳,表明积分补偿的加入能够有效消除位置跟踪静差,提高控制的准确性,增强系统的稳定性。To realize the autonomous control of the position and attitude of quadrotor,the intelligent controller was designed in combination with the deep deterministic policy gradient(DDPG)algorithm in reinforcement learning.However,the controller suffered from the steady-state error during position tracking.The cause of this problem was analyzed in detail.To deal with the steady-state error,the DDPG algorithm with integral compensator was proposed.The simulation results show that the controller based on DDPG with integral compensator is able to realize the stable control of quadrotor by self-learning and the position error is reduced to zero.Besides,the control process remains stable after the reward function changes.The results demonstrate that the addition of integral compensation effectively eliminates the steady-state error of positon control and improves the accuracy of controller.Moreover,the integral compensator enhances the stability of the system.

关 键 词：深度确定性策略梯度 积分补偿 静差 四旋翼控制 自主学习 智能控制 未知动力学模型 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]