Sei sulla pagina 1di 151

UNIVERSITY OF CALIFORNIA SAN DIEGO

Adaptiv e Global Optimization with Local Search

A dissertation submitted in partial satisfaction of the

requirements for the degree

in Computer Science

Doctor of Philosophy

Engineering

by

William Eugene

Hart

Committee in charge

Professor

Richard K Belew Chair

Professor

Philip E Gill

Professor

Paul R Kube

Professor

Christos H Papadimitriou

Professor

J Benjamin Rosen

Professor

Halbert L White Jr

Cop yright

William Eugene Hart

All rights reserved

The dissertation of William Eugene Hart is approved

and it is acceptable in quality and form for publication

on micro lm

Chair

University of California San Diego

iii

T ABLE OF CONTENTS

Signature Page

iii

Dedication

iv

Table of Contents

v

List

of

Figures

viii

List

of

Tables

 

x

Vita and Publications

xii

Abstract

xiv

I Introduction

A Global Optimization

B Adaptive Global Optimization

C Genetic Algorithms with Local Search

D Parallel Genetic Algorithms with Local Search

E

Dissertation Overview

II Background and Related Work

A Local Search

 

Random Local Search

Conjugate Gradient

 

Stochastic

Approximation

B Global Optimization

Methods with Guaranteed Accuracy

Indirect Methods

Direct Methods

C Evolutionary Search

Genetic Algorithms

Panmictic and Geographically Structured Genetic Algorithms

GAs with Local Search

 

D Related Work

III Local Search with Nonadaptive Global Search

A De nitions

B Complexity Analysis

C Monte Carlo vs Multistart

Monte Carlo Complexity

Multistart Complexity

Comparison

 

v

D

Probabilistic Multistart

 

Complexity

Comparison

E

Summary

IV Test Problems and Methods

A Worst Case Analysis

 

Complexity Analysis

Performance Guarantees

B Test Problems

 

 

Griewank

Modi ed Griewank

Rastrigin

C Optimization Methods

 

Floating

Point

GA

Floating Point

Mutation

Performance Comparisons

V Selective Local Search

A Introduction

 

B Nonadaptive Selection

 

Fixed Frequency Local Search

Elitism

Population Size and Local Search Length

C Adaptive Selection

Distribution base Adaptation

Fitness based Adaptation

Results

Discussion

D Summary and Discussion

VI Parallel Geographically Structured Genetic Algorithms

A Introduction

 

B A MIMD GSGA Design

C Complexity Analysis I

Time Complexity

 

Performance Analysis

D Complexity Analysis II

 

Local Search on Random Subsets

Variable Length

Local Search

E Methods

F Experimental Results