基于共轭梯度法的改进型BP神经网络PID控制算法

摘要

To deal with the defects of the classic BP neural network PID controlor in slowly converging and easily immerging in partial minimum, this paper proposes a improved BP neural network PID controlor based on conjugate gradient, uses Polak-Ribiere linear search methods, to improve the classic BP neural network PID controlor, can make the training of neural network faster and can eliminate immerging in partial minimum. The simulation results of Matlab programs show that this way is effective.%针对常规BP神经网络PID控制器存在收敛速度慢和易陷入局部极小的缺点,提出一种基于共轭梯度算法的改进型BP神经网络PID控制器,采用Polak -Ribiere线性搜索方法,对传统BP神经网络PID控制器进行改进,加快了网络训练速度,避免网络陷入局部极小.在Matlab平台下实现算法程序,仿真结果表明该改进控制方法的有效性.