广义PID实时递归学习算法  

The Generalized PID Real Time Recurrent Learning Algorithm

作　　者：张俊杰[1] 杨艳丽[1] 尤昌德[1] 

机构地区：[1]西安交通大学

出　　处：《西安交通大学学报》1997年第1期104-110,共7页Journal of Xi'an Jiaotong University

摘　　要：利用经典的ＰＩＤ控制思想，在实时递归学习（ＲＴＲＬ）算法基础上，研究了微分项和积分项对目标函数收敛行为的影响．结果表明：只要能动态地调整积分项和微分项，收敛速度就会大大加快且能跨越局部极小值．基于此，提出了改进ＲＴＲＬ算法，包括ＰＩＤ参数自动调整和积分项、微分项动态调整规则．采用遗忘因子有效地解决了过去状态对当前灵敏度过度影响的问题．通过一阶混沌序列预测的仿真表明，文中的算法较之标准ＲＴＲＬ和动量项ＲＴＲＬ算法，有更快的学习收敛速度。The real time recurrent learning(RTRL) algorithm is modefied by using the classical proportional integral derivative (PID) control scheme. The convergence is faster and the local minimum can be overcome in the new algorithm. The effect of the integral and derivation terms on the covergence behavior of the objective function is studied and a set of principles is presented to appropriately tune the PID parameters. A forgetting factor is used for the sensitivities of the PID RTRL in order to decrease the too much effect of the past states. With the simulation of a first order chaotic sequence, it is observed that the rate of covergence is greatly improved and the local minimum can be overcome.

关 键 词：PID检测 学习算法 实时递归学习 递归神经网纲 

分 类 号：TP273.22[自动化与计算机技术—检测技术与自动化装置]