用于动力系统控制的时滞神经网络  

Time delay neural network for dynamical system control

作　　者：徐旭[1] 葛宏伟[2] 梁艳春[3] 

机构地区：[1]吉林大学数学学院,长春130012 [2]大连理工大学电子与信息工程学院,辽宁大连116024 [3]吉林大学计算机科学与技术学院,长春130012

出　　处：《吉林大学学报（工学版）》2009年第4期1047-1051,共5页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(10501017;60673023)

摘　　要：提出了一种连续回归时滞神经元网络,并将其用于动力系统控制。通过在隐层神经元中引入时滞,构造了一种新的时滞神经网络模型。应用Lyapunov稳定性方法,给出了此神经网络权值的修正算法和系统的控制输入,保证了控制系统的稳定性。本文建立的控制方法不需要系统任何的先验知识,并能保证控制误差收敛到零点附近的一个小的邻域,并始终保持在这个区域内。最后通过对一种超声马达进行控制实验,验证了本方法的有效性。A novel time delay neural network for dynamical system control was presented. By introducing time delay into the hidden neurons, a new neural network model with time delay was constructed. On the basis of Lyapunov stability approach, the training algorithm for determining the weights in neural network was developed and the control laws that ensure the stability of the control system were also developed. Apriori knowledge is not required in the proposed control method, and the control error converges to a small range near the zero point and remains within the range throughout the execution. Numerical control experiment for an ultrasonic motor was performed and the result validates the proposed algorithm.

关 键 词：人工智能 稳定性 时滞 神经网络 动力系统控制 超声马达 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]