基于改进BP神经网络的PID控制方法研究  

Study of PID Control Based on Improved BP Neural Network

作　　者：史春朝[1] 张国山[1] 

机构地区：[1]天津大学电气与自动化工程学院,天津300072

出　　处：《计算机仿真》2006年第12期156-159,共4页Computer Simulation

摘　　要：针对最速下降法收敛速度慢和易陷入局部极小的缺点,提出一种新型的基于改进BP神经网络的PID控制方法,该方法将神经网络和PID控制策略相结合,既具有神经网络自学习、自适应及逼近任意函数的能力,又具有常规PID控制器结构简单的特点。该控制器的算法采用F letcher-R eeves共轭梯度法,它可以避免网络陷入局部极小点,同时加快网络的训练速度。并用这种改进的共轭梯度法对神经网络PID控制器参数实现在线修正。最后给出了在M atlab平台上的实现算法,仿真结果表明该控制方法是有效的。To deal with the defects of the steepest descent in slowly converging and easily immerging in partial minimum, this paper proposes a new type of PID control method based on the improved BP neural network, which is a combination of the neural network and the PID strategy. It has the merits of both neural network and PID controller. Moreover, Fletcher - Reeves conjugate gradient in controller can make the training of network faster and can eliminate the disadvantages of steepest descent in BP algorithnL The parameters of the neural network PID controller are modified on line by the improved conjugate gradient. The programming steps under MATLAB are finally described. Simulation result shows that the controller is effective.

关 键 词：误差反传神经网络 改进共轭梯度法 比例-积分-微分控制 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]