基于Q-MPC的径向四自由度磁悬浮轴承控制策略  

Control Strategy of Radial Four Degree of Freedom Magnetic Bearings Based on Q-MPC

作　　者：邹晋彬 易辉阳 张鹏辉 冯悦昕 邓自刚 ZOU Jinbin;YI Huiyang;ZHANG Penghui;FENG Yuexin;DENG Zigang(State Key Laboratory of Rail Transit Vehicle System,Chengdu 610031,China;The School of Information Science&Technology,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]轨道交通运载系统全国重点实验室,成都610031 [2]西南交通大学信息科学与技术学院,成都610031

出　　处：《轴承》2024年第7期36-44,共9页Bearing

摘　　要：针对磁悬浮轴承系统的高度非线性、不稳定性以及多输入多输出特点,将强化学习Q-Learning算法同模型预测控制(MPC)结合,设计了一种根据当前系统状态自适应整定参数的模型预测整体控制器(Q-MPC)。以径向四自由度磁悬浮轴承转子位移为控制对象,首先建立径向四自由度磁悬浮轴承数学模型,在此基础上引入希尔德雷斯二次规划方法约束并优化MPC控制器输入,利用Q-Learning算法整定MPC控制时域和预测时域,得到适用于径向四自由度磁悬浮轴承系统的Q-MPC控制器。仿真结果表明:正弦参考信号下,Q-MPC控制器作用的系统各自由度平均跟踪误差相比MPC控制器降低了36%;恒定参考信号下,相比于MPC控制器,Q-MPC控制器作用的系统各自由度的平均超调量降低了51.4%,系统稳定所需平均调整时间降低了14.4%,施加扰动后,4个自由度的平均波动幅度降低了76.1%,系统稳定所需平均调整时间降低了26.2%,验证了所设计的Q-MPC控制器的有效性。Aimed at high nonlinearity,instability and multi-input multi-output characteristics of magnetic bearing system,Q-MPC is designed by combining reinforcement learning Q-Learning algorithm with model predictive control(MPC)to tune the parameters adaptively based on current system state.Taking rotor displacement of radial four degree of freedom magnetic bearings as control object,a mathematical model of radial four degree of freedom magnetic bearings is established firstly.Based on this,Hildreth’s quadratic programming procedure is introduced to constrain and optimize the input of MPC controller.Q-Learning algorithm is used to tune the MPC control time domain and predictive time domain,and Q-MPC controller suitable for radial four degree of freedom magnetic bearing system is obtained.The simulation results show that under sinusoidal reference signal,compared with MPC controller,the tracking error of each degree of freedom of the system acted by Q-MPC controller is reduced by an average of 36%;under constant reference signal,compared with MPC controller,the overshoot of each degree of freedom of the system acted by Q-MPC controller is reduced by an average of 51.4%,and the average time required for system stability is reduced by 14.4%,after disturbance is applied,the fluctuation amplitude of four degrees of freedom is reduced by an average of 76.1%,and the average time required for system stability is reduced by 26.2%,verifying the effectiveness of designed Q-MPC controller.

关 键 词：滑动轴承 磁力轴承 径向轴承 模型预测控制 强化学习 

分 类 号：TH133.31[机械工程—机械制造及自动化] O313.7[理学—一般力学与力学基础]