基于RBF神经网络控制的球杆系统位置控制实验研究*

球杆系统是一种典型的高阶非线性不稳定系统,针对PID跟踪控制精度不高及BP神经网络控制训练时间较长的问题,本文提出一种带有低通滤波器的RBF神经网络控制器(RBFC)动态补偿PID控制的球杆控制方法,控制系统由RBF神经网络控制及PID控制器组成。为提高参数辨识速度和避免局部最小值,采用梯度下降法更新隐含层参数,采用带有遗忘因子的最小二乘法更新输出层权值。实验结果表明,该控制方案相比PID控制具有更高的控制精度,比BP神经网络具有更快的学习速度,低通滤波器保证了RBFC的辨识精度和稳定的控制输出,具有良好的动静态特性和控制性能。

Experimental study of ball and beam system position control based on RBF neural network control

Considering the poor control accuracy of PID control and long training time of BP neural network in ball-and-beam system position control experiment, this paper proposed a new control method based RBF neural network dynamically compensating PID control with lowpass-filter, the control system was composed of RBF neural network controller and PID controller. For improving the speed of parameter identification and avoiding the local minimum value, this paper used the gradient descent algorithm to update implicit-layer parameters, and used the least square method to adjust output-layer weights with the forgetting factor. The experimental results show that this control method has higher accuracy than PID controller and faster learning speed than BP neural network controller, the lowpass-filter ensure the good identification precision and stable output of control system, this method has good control performance and excellent dynamic-and-static character.