Parameter identification via neural networks with fast convergence

被引：16

作者：

Yadaiah, N ^{[1
]}

Sivakumar, L

Deekshatulu, BL

机构：

[1] JNT Univ, Hyderabad 500028, Andhra Pradesh, India

[2] BHEL R&D, Hyderabad 500093, Andhra Pradesh, India

[3] CSSTE AP, Dehra Dun 248001, Uttar Pradesh, India

来源：

MATHEMATICS AND COMPUTERS IN SIMULATION | 2000年 / 51卷 / 3-4期

关键词：

artificial neural networks; parameter identification; optimization; supervised learning; performance index;

D O I：

10.1016/S0378-4754(99)00114-7

中图分类号：

TP39 [计算机的应用];

学科分类号：

081203 ; 0835 ;

摘要：

The parameter identification using artificial neural networks is becoming very popular. In this chapter, the parameters of dynamical system are identified using artificial neural networks. A fast gradient decent technique for the parameter identification of a linear dynamical system has been presented. The following concepts are used for training of neural networks while identifying the system parameters: (1) batch wise training of neural networks; (2) variable learning parameter and; (3) an intelligent check over the rate at which parameters are converging. The complete algorithm is summarized as a flow chart. A detailed mathematical formulation is given. The simulation results and a comparative study with existing method is included. (C) 2000 IMACS/Elsevier Science B.V. All rights reserved.

引用

页码：157 / 167

页数：11

共 15 条

[1] SINGLE-LAYER NEURAL NETWORKS FOR LINEAR-SYSTEM IDENTIFICATION USING GRADIENT DESCENT TECHNIQUE [J].