Click here to load reader

LNCS 3102 - Robust and Efficient Genetic Algorithms with Hierarchical Niching · PDF file 2005-06-28 · Robust and Efficient Genetic Algorithms with Hierarchical Niching 1221. In

  • View
    0

  • Download
    0

Embed Size (px)

Text of LNCS 3102 - Robust and Efficient Genetic Algorithms with Hierarchical Niching · PDF file...

  • K. Deb et al. (Eds.): GECCO 2004, LNCS 3102, pp. 1220–1232, 2004. © Springer-Verlag Berlin Heidelberg 2004

    Robust and Efficient Genetic Algorithms with Hierarchical Niching and a Sustainable Evolutionary

    Computation Model

    Jianjun Hu1 and Erik Goodman2

    1,2Genetic Algorithm Research and Application Group( GARAGe) 1Department of Computer Science and Engineering

    2Department of Electrical and Computer Engineering Michigan State University, East Lansing, MI, 48823

    {Hujianju, Goodman}@egr.msu.edu

    Abstract. This paper proposes a new niching method named hierarchical niching, which combines spatial niching in search space and a continuous tem- poral niching concept. The method is naturally implemented as a new genetic algorithm, QHFC, under a sustainable evolutionary computation model: the Hi- erarchical Fair Competition (HFC) Model. By combining the benefits of the temporally continuing search capability of HFC and this spatial niching capa- bility, QHFC is able to achieve much better performance than deterministic crowding and restricted tournament selection in terms of robustness, efficiency, and scalability, simultaneously, as demonstrated using three massively multi- modal benchmark problems. HFC-based genetic algorithms with hierarchical niching seem to be very promising for solving difficult real-world problems.

    1 Introduction

    Genetic algorithms are widely applied to challenging engineering problems today. However, there are still several undesirable properties with current genetic algo- rithms. The first one is the lack of a quality guarantee of genetic search. For example, genetic algorithms are usually sensitive to the population size in terms of their search capability. Unfortunately, it is difficult to estimate the required population size, de- spite the extant population sizing theory [1,2]. Too large a population size leads to low efficiency, and one that is too small may simply fail to achieve satisfactory re- sults. The second undesirable property is that once a genetic algorithm stagnates dur- ing a search, it usually loses most of its search capability, and there is no good way to rejuvenate the run in an efficient manner. Simple restart or strong mutations may waste the computations spent before by destroying the building blocks in the popula- tion. Weak mutations may perturb the solutions a little bit, but they cannot incur sig- nificant move in search space once the framework of the individual is established. The third problem of current genetic algorithms is the lack of robustness such as large variation of the performance of several runs due to the opportunistic and convergent nature of current genetic algorithms.

    Verwendete Distiller 5.0.x Joboptions Dieser Report wurde automatisch mit Hilfe der Adobe Acrobat Distiller Erweiterung "Distiller Secrets v1.0.5" der IMPRESSED GmbH erstellt. Sie koennen diese Startup-Datei für die Distiller Versionen 4.0.5 und 5.0.x kostenlos unter http://www.impressed.de herunterladen.

    ALLGEMEIN ---------------------------------------- Dateioptionen: Kompatibilität: PDF 1.3 Für schnelle Web-Anzeige optimieren: Nein Piktogramme einbetten: Nein Seiten automatisch drehen: Nein Seiten von: 1 Seiten bis: Alle Seiten Bund: Links Auflösung: [ 2400 2400 ] dpi Papierformat: [ 595 842 ] Punkt

    KOMPRIMIERUNG ---------------------------------------- Farbbilder: Downsampling: Ja Berechnungsmethode: Bikubische Neuberechnung Downsample-Auflösung: 300 dpi Downsampling für Bilder über: 450 dpi Komprimieren: Ja Automatische Bestimmung der Komprimierungsart: Ja JPEG-Qualität: Maximal Bitanzahl pro Pixel: Wie Original Bit Graustufenbilder: Downsampling: Ja Berechnungsmethode: Bikubische Neuberechnung Downsample-Auflösung: 300 dpi Downsampling für Bilder über: 450 dpi Komprimieren: Ja Automatische Bestimmung der Komprimierungsart: Ja JPEG-Qualität: Maximal Bitanzahl pro Pixel: Wie Original Bit Schwarzweiß-Bilder: Downsampling: Ja Berechnungsmethode: Bikubische Neuberechnung Downsample-Auflösung: 2400 dpi Downsampling für Bilder über: 24000 dpi Komprimieren: Ja Komprimierungsart: CCITT CCITT-Gruppe: 4 Graustufen glätten: Nein

    Text und Vektorgrafiken komprimieren: Ja

    SCHRIFTEN ---------------------------------------- Alle Schriften einbetten: Ja Untergruppen aller eingebetteten Schriften: Nein Wenn Einbetten fehlschlägt: Warnen und weiter Einbetten: Immer einbetten: [ /Courier-BoldOblique /Helvetica-BoldOblique /Courier /Helvetica-Bold /Times-Bold /Courier-Bold /Helvetica /Times-BoldItalic /Times-Roman /ZapfDingbats /Times-Italic /Helvetica-Oblique /Courier-Oblique /Symbol ] Nie einbetten: [ ]

    FARBE(N) ---------------------------------------- Farbmanagement: Farbumrechnungsmethode: Farbe nicht ändern Methode: Standard Geräteabhängige Daten: Einstellungen für Überdrucken beibehalten: Ja Unterfarbreduktion und Schwarzaufbau beibehalten: Ja Transferfunktionen: Anwenden Rastereinstellungen beibehalten: Ja

    ERWEITERT ---------------------------------------- Optionen: Prolog/Epilog verwenden: Ja PostScript-Datei darf Einstellungen überschreiben: Ja Level 2 copypage-Semantik beibehalten: Ja Portable Job Ticket in PDF-Datei speichern: Nein Illustrator-Überdruckmodus: Ja Farbverläufe zu weichen Nuancen konvertieren: Ja ASCII-Format: Nein Document Structuring Conventions (DSC): DSC-Kommentare verarbeiten: Ja DSC-Warnungen protokollieren: Nein Für EPS-Dateien Seitengröße ändern und Grafiken zentrieren: Ja EPS-Info von DSC beibehalten: Ja OPI-Kommentare beibehalten: Nein Dokumentinfo von DSC beibehalten: Ja

    ANDERE ---------------------------------------- Distiller-Kern Version: 5000 ZIP-Komprimierung verwenden: Ja Optimierungen deaktivieren: Nein Bildspeicher: 524288 Byte Farbbilder glätten: Nein Graustufenbilder glätten: Nein Bilder (< 257 Farben) in indizierten Farbraum konvertieren: Ja sRGB ICC-Profil: sRGB IEC61966-2.1

    ENDE DES REPORTS ----------------------------------------

    IMPRESSED GmbH Bahrenfelder Chaussee 49 22761 Hamburg, Germany Tel. +49 40 897189-0 Fax +49 40 897189-71 Email: [email protected] Web: www.impressed.de

    Adobe Acrobat Distiller 5.0.x Joboption Datei

    /AutoFilterColorImages true /sRGBProfile (sRGB IEC61966-2.1) /ColorImageDepth -1 /PreserveOverprintSettings true /AutoRotatePages /None /UCRandBGInfo /Preserve /EmbedAllFonts true /CompatibilityLevel 1.3 /StartPage 1 /AntiAliasColorImages false /CreateJobTicket false /ConvertImagesToIndexed true /ColorImageDownsampleType /Bicubic /ColorImageDownsampleThreshold 1.5 /MonoImageDownsampleType /Bicubic /DetectBlends true /GrayImageDownsampleType /Bicubic /PreserveEPSInfo true /GrayACSImageDict > /ColorACSImageDict > /PreserveCopyPage true /EncodeMonoImages true /ColorConversionStrategy /LeaveColorUnchanged /PreserveOPIComments false /AntiAliasGrayImages false /GrayImageDepth -1 /ColorImageResolution 300 /EndPage -1 /AutoPositionEPSFiles true /MonoImageDepth -1 /TransferFunctionInfo /Apply /EncodeGrayImages true /DownsampleGrayImages true /DownsampleMonoImages true /DownsampleColorImages true /MonoImageDownsampleThreshold 10.0 /MonoImageDict > /Binding /Left /CalCMYKProfile (U.S. Web Coated (SWOP) v2) /MonoImageResolution 2400 /AutoFilterGrayImages true /AlwaysEmbed [ /Courier-BoldOblique /Helvetica-BoldOblique /Courier /Helvetica-Bold /Times-Bold /Courier-Bold /Helvetica /Times-BoldItalic /Times-Roman /ZapfDingbats /Times-Italic /Helvetica-Oblique /Courier-Oblique /Symbol ] /ImageMemory 524288 /SubsetFonts false /DefaultRenderingIntent /Default /OPM 1 /MonoImageFilter /CCITTFaxEncode /GrayImageResolution 300 /ColorImageFilter /DCTEncode /PreserveHalftoneInfo true /ColorImageDict > /ASCII85EncodePages false /LockDistillerParams false >> setdistillerparams > setpagedevice

  • Robust and Efficient Genetic Algorithms with Hierarchical Niching 1221

    In the past three decades, many niching techniques have been proposed, which have greatly improved the scalability and robustness of genetic search for difficult multi-modal problems [3]. However, due to the convergent nature of the current ge- netic algorithm framework, these niching approaches still meet difficulty in many hard problems. Based on a sustainable evolutionary computation framework and a hi- erarchical niching mechanism, this paper proposes a new genetic algorithm, named QHFC, which can significantly improve robustness, efficiency and scalability over that of a representative modern niching approach.

    The rest of the paper is organized as follows. In section 2, existing commonly used niching techniques including temporal niching and spatial niching are surveyed, and their three inherent difficulties are outlined. Section 3 then presents the ideas of the sustainable evolutionary computation framework of HFC [4,5], which underlies the design of a new genetic algorithm with hierarchical niching, QHFC to be described in Section 4. A set of three well-known genetic algorithm benchmark problems are used to evaluate QHFC in section 5 and the results are compared to genetic algorithms with deterministic crowding and restricted tournament selection in terms of scalability, ef- ficiency, and robustness. A conclusion is then drawn

Search related