基于神经网络反馈补偿控制的磁悬浮球位置控制  

Magnetic levitation ball position control based on neural network feedback compensation control

作　　者：朱坚民[1] 沈正强[1] 李孝茹[1] 齐北川 

机构地区：[1]上海理工大学机械工程学院,上海200093

出　　处：《仪器仪表学报》2014年第5期976-986,共11页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(50975179);上海市教委科研创新项目(11ZZ136);上海市科委科研计划(11DZ0511400);上海市科委科研计划(12DZ2252300)资助项目

摘　　要：针对现有磁悬浮球位置控制算法控制精度不高的问题,提出了一种基于神经网络反馈补偿控制、PID控制和神经网络辨识器的磁悬浮球位置控制结构。采用神经网络辨识器建立控制系统误差与控制量的动态模型,在线学习时,神经网络辨识器在磁悬浮球位置实时控制过程中被训练,其训练后的参数动态复制给神经网络反馈补偿控制器,避免了神经网络反馈补偿控制器的离线训练。神经网络反馈补偿控制器在实时控制过程中基于PID的控制输出作进一步的学习,以补偿神经网络辨识器的建模误差并产生一个反馈补偿控制量。在控制回路中引入PID控制器可以保证神经网络在学习初期或系统受到扰动时闭环系统的稳定性,提高控制算法的鲁棒性。仿真与实验结果表明:神经网络反馈补偿控制方法较PID控制方法的控制精度由0.5mm提高到0.04 mm,控制系统具有良好的动静态性能和较强的鲁棒性。Aiming at the low control accuracy problem of present magnetic levitation ball position control algorithms,a control structure of magnetic levitation ball position based on neural network feedback compensation control,PID control and neural network identifier is proposed in this paper.The neural network identifier is adopted to establish the dynamic model of the error and control quantity of the control system.During online learning,the neural network identifier is trained in the real-time control process of magnetic levitation ball position; after the training,the training parameters are dynamically replicated to the neural network feedback compensation controller,which avoids the offline training of the neural network feedback compensation controller.During the real-time control process,the neural network feedback compensation controller could further study based on the control output of the PID controller,which compensates the modeling error of the neural network identifier and produces a feedback compensation control quantity.The introduction of the PID controller in the control loop can guarantee the stability of the close loop system at the initial stage of the neural network learning or when the system is disturbed,which improves the robustness of the control algorithm.The simulation and experiment results indicate that the neural network feedback compensation control method could achieve a control precision of 0.04 mm,which is better than the control precision of 0.5 mm for PID control method.The control system shows excellent static and dynamic performance and remarkable robustness.

关 键 词：磁悬浮球 神经网络 补偿控制 系统辨识 位置控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] TH89[自动化与计算机技术—控制科学与工程]