An analysis of a class of neural networks for solving linear programming problems

被引:98
作者
Chong, EKP [1 ]
Hui, S
Zak, SH
机构
[1] Purdue Univ, Sch Elect & Comp Engn, W Lafayette, IN 47907 USA
[2] San Diego State Univ, Dept Math Sci, San Diego, CA 92182 USA
基金
美国国家科学基金会;
关键词
invariant sets; linear programming; neural networks; nondifferentiable optimization; penalty functions; sliding modes;
D O I
10.1109/9.802909
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
A class of neural networks that solve linear programming problems is analyzed, The neural networks considered are modeled bq dynamic gradient systems that are constructed using a parametric family of exact (nondifferentiable) penalty functions. It is pro, ed that for a given linear programming problem and sufficiently large penalty parameters, any trajectory of the neural network converges in finite time to its solution set, For the analysis, Lyapunov-type theorems are developed for finite time convergence of nonsmooth sliding mode dynamic systems to invariant sets, The results are illustrated via numerical simulation examples.
引用
收藏
页码:1995 / 2006
页数:12
相关论文
共 36 条
[1]  
[Anonymous], 1985, NONDIFFERENTIABLE OP
[2]  
[Anonymous], 1988, Real analysis
[3]  
[Anonymous], APPL MATH PROGRAMMIN
[4]  
Aubin J.-P., 1984, DIFFERENTIAL INCLUSI
[5]   NECESSARY AND SUFFICIENT CONDITIONS FOR A PENALTY METHOD TO BE EXACT [J].
BERTSEKAS, DP .
MATHEMATICAL PROGRAMMING, 1975, 9 (01) :87-99
[6]  
CICHOCKI A, 1993, NEURAL
[7]   LINEAR-PROGRAMMING VIA A NON-DIFFERENTIABLE PENALTY FUNCTION [J].
CONN, AR .
SIAM JOURNAL ON NUMERICAL ANALYSIS, 1976, 13 (01) :145-154
[8]   VARIABLE STRUCTURE CONTROL OF NONLINEAR MULTIVARIABLE SYSTEMS - A TUTORIAL [J].
DECARLO, RA ;
ZAK, SH ;
MATTHEWS, GP .
PROCEEDINGS OF THE IEEE, 1988, 76 (03) :212-232
[9]  
Drazenovic B., 1969, Automatica, V5, P287, DOI 10.1016/0005-1098(69)90071-5
[10]  
Fang S.-C., 1993, Linear Optimization and Extensions: Theory and Algorithms, VFirst