基于神经网络延时预测的自适应网络控制系统  

Adaptive networked control system based on delay prediction using neural network

作　　者：于晓明[1] 蒋静坪[1] 

机构地区：[1]浙江大学电气工程学院,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2012年第2期194-198,231,共6页Journal of Zhejiang University：Engineering Science

基　　金：国家博士点学科专项科研基金资助项目(20030335002);浙江省科技厅资助项目(2004C31084)

摘　　要：针对网络控制系统存在着随机、时变、不确定的信息传输延时,采用带有时间戳的线性神经网络(TSLNN)进行在线延时预测,实时地获得当前采样周期的网络传输延时预测值.该方法选取3个先验的网络实测延时值作为神经网络的输入样本,选用widrow-hoff学习规则作为神经网络的训练算法;应用网络传输延时预测值,并采用一阶Pade方法,对数学模型中的延时环节进行线性化处理,从而获得无刷直流电机调速网络控制系统的线性数学模型;最后,利用模型参考自适应控制方法(MRAC)设计闭环控制器.仿真结果表明,将基于TSLNN在线延时预测的MRAC方法应用于无刷直流电机调速网络控制系统中,可以获得令人满意的系统动、静态性能.Aiming at existent randomness,time varying and uncertainty of the information transmission time delay in the networked control systems,the time-stamped linear neural network（TSLNN） is adopted to predict the time delay in real-time,on line.Using measurement time delay values of the previous sampled period in actual network as the input data set for the neural network,the widrow-hoff learning rule is chosen as the training algorithm of neural network;With predicted network delay value,and using one-order Pade method to linearize the delay element,the linear mathematical model of the brushless direct-current motor drive networked control systems is established;The model reference adaptive control（MRAC） strategy is introduced to design the close-loop controller for the control system;The results of digital simulation prove that based on TSLNN time delay prediction in real-time,the MRAC for the brushless direct-current motor drive networked control system is feasible,and the dynamic and static response performances of the system are satisfied.

关 键 词：带有时间戳的线性神经网络(TSLNN) 网络控制系统 一阶Pade方法 无刷直流电机调速系统 模型参考自适应控制(MRAC) 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]