Bánhelyi / Csendes / Lévai The GLOBAL Optimization Algorithm

Newly Updated with Java Implementation and Parallelization
1. Auflage 2018
ISBN: 978-3-030-02375-1
Verlag: Springer International Publishing
Format: PDF
Kopierschutz: 1 - PDF Watermark

E-Book, Englisch, 112 Seiten

Reihe: SpringerBriefs in Optimization

This book explores the updated version of the GLOBAL algorithm which contains improvements for a local search algorithm and new Java implementations. Efficiency comparisons to earlier versions and on the increased speed achieved by the parallelization, are detailed. Examples are provided for students as well as researchers and practitioners in optimization, operations research, and mathematics to compose their own scripts with ease. A GLOBAL manual is presented in the appendix to assist new users with modules and test functions.  

GLOBAL is a successful stochastic multistart global optimization algorithm that has passed several computational tests, and is efficient and reliable for small to medium dimensional global optimization problems. The algorithm uses clustering to ensure efficiency and is modular in regard to the two local search methods it starts with, but though it can easily apply other local techniques. The strength of this algorithm lays lies in its reliability and adaptive algorithm parameters. The GLOBAL algorithm is free to download in the earlier Fortran, C, and MATLAB implementations.



 

Autoren/Hrsg.

Bánhelyi, Balázs

Csendes, Tibor

Lévai, Balázs

Pál, László

Zombori, Dániel

Inhaltsverzeichnis

1;Acknowledgments;5
2;Contents;6
3;1 Introduction;9
3.1;1.1 Introduction;9
3.2;1.2 Problem Domain;10
3.3;1.3 The GLOBAL Algorithm;10
4;2 Local Search;14
4.1;2.1 Introduction;14
4.2;2.2 Local Search Algorithms;15
4.2.1;2.2.1 Derivative-Free Local Search;15
4.2.2;2.2.2 The Basic UNIRANDI Method;16
4.2.3;2.2.3 The New UNIRANDI Algorithm;16
4.2.4;2.2.4 Reference Algorithms;21
4.3;2.3 Computational Investigations;22
4.3.1;2.3.1 Experimental Settings;22
4.3.2;2.3.2 Comparison of the Two UNIRANDI Versions;23
4.3.3;2.3.3 Comparison with Other Algorithms;25
4.3.4;2.3.4 Error Analysis;26
4.3.5;2.3.5 Performance Profiles;29
4.4;2.4 Conclusions;32
5;3 The GLOBALJ Framework;33
5.1;3.1 Introduction;33
5.2;3.2 Switching from MATLAB to JAVA;34
5.3;3.3 Modularization;34
5.4;3.4 Algorithmic Improvements;37
5.5;3.5 Results;43
5.6;3.6 Conclusions;45
6;4 Parallelization;46
6.1;4.1 Introduction;46
6.2;4.2 Parallel Techniques;47
6.2.1;4.2.1 Principles of Parallel Computation;47
6.3;4.3 Design of PGLOBAL Based on GLOBAL;49
6.4;4.4 Implementation of the PGlobal Algorithm;53
6.4.1;4.4.1 SerializedGlobal;53
6.4.2;4.4.2 SerializedClusterizer;56
6.5;4.5 Parallelized Local Search;61
6.6;4.6 Losses Caused by Parallelization;61
6.7;4.7 Algorithm Parameters;61
6.8;4.8 Results;62
6.8.1;4.8.1 Environment;62
6.8.2;4.8.2 SerializedGlobal Parallelization Test;63
6.8.3;4.8.3 SerializedGlobalSingleLinkageClusterizer Parallelization Test;66
6.8.4;4.8.4 Comparison of Global and PGlobal Implementations;67
6.9;4.9 Conclusions;71
7;5 Example;73
7.1;5.1 Environment;73
7.2;5.2 Objective Function;73
7.3;5.3 Optimizer Setup;75
7.4;5.4 Run the Optimizer;76
7.5;5.5 Constraints;77
7.6;5.6 Custom Module Implementation;81
8;Appendix A User's Guide;84
8.1;A.1 Global Module;84
8.1.1;A.1.1 Parameters;84
8.2;A.2 SerializedGlobal Module;85
8.2.1;A.2.1 Parameters;85
8.3;A.3 GlobalSingleLinkageClusterizer Module;86
8.3.1;A.3.1 Parameters;86
8.4;A.4 SerializedGlobalSingleLinkageClusterizer Module;87
8.4.1;A.4.1 Parameters;87
8.5;A.5 UNIRANDI Module;87
8.5.1;A.5.1 Parameters;87
8.6;A.6 NUnirandi Module;88
8.6.1;A.6.1 Parameters;88
8.7;A.7 UnirandiCLS Module;88
8.7.1;A.7.1 Parameters;89
8.8;A.8 NUnirandiCLS Module;89
8.8.1;A.8.1 Parameters;89
8.9;A.9 Rosenbrock Module;90
8.9.1;A.9.1 Parameters;90
8.10;A.10 LineSearchImpl Module;90
9;Appendix B Test Functions;91
10;Appendix C DiscreteClimber Code;102
11;References;108