伪线性神经网络  

A New MIMO Pseudolinear Neural Network

作　　者：李玉云[1] 王秋庭[1] 

机构地区：[1]武汉科技大学城市建设学院,湖北武汉430070

出　　处：《武汉科技大学学报》2002年第1期77-80,共4页Journal of Wuhan University of Science and Technology

基　　金：武汉市建委 2 0 0 1年建设科技基金资助项目

摘　　要：给出了多输入 多输出 (MIMO)的伪线性神经网络 (PNN )结构与模型。PNN的构成是基于Gain scheduling控制技术与扩展线性化方法。还讨论PNN的递推预报误差的二阶学习算法 (RPF)和训练过程 ,并通过对多变量、强耦合的非线性加热炉的动态建模 ,证明了用PNN建模 ,效果良好 ,且具有训练参数少 ,收敛速度快 。A novel MIMO pseudolinear neural network (MIMOPNN) for modeling of nonlinear dynamic control system is presented, which,based on the Gain scheduling control techniques and extended linearization of nonlinear system,possesses multilayered architecture. The input of MIMOPNN is composed of inputs and outputs of system;the output layer is with form of weighting multiplication of input and output of the hidden layer, possessing a quasi linear form of describing the dynamic characteristics of system. A recursive prediction error learning algorithm is adopted for training the network. The application of this method to dynamic modeling of MIMO nonlinear heating furnace shows that the new type network is satisfactory.

关 键 词：伪线性神经网络 非线性系统 RFP 系统建模 加热炉 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]