ICSPEA: Integrating CMA-ES and SPEA2

Mehnen, J.1, a; Wagner, T.1, b; Kersting, P.1, c; Tipura, I.; Rudolph, G.2

1)
Institut für Spanende Fertigung, Universität Dortmund, Baroper Str. 301, 44227 Dortmund
2)
Universität Dortmund, Fakultät für Informatik, Lehrstuhl für Algorithm Engineering (LS 11), 44221 Dortmund

a) mehnen@isf.de; b) wagner@isf.de; c) petra.kersting@isf.de

Kurzfassung

The covariance matrix adaptation (CMA) is a concept originally introduced for improving the single-objective evolution strategy (ES). CMA varies the classical ES-mutation operator by utilising a mutation distribution adaptation scheme and an evolution path, which takes the evolutionary history into account. SPEA2 surely belongs to the most popular multi-objective evolutionary algorithms. It uses the strength Pareto concept and a special distribution measure for the evaluation of offspring individuals. An archive collects non-dominated individuals, which are used during selection, making the SPEA2 a typical elitist strategy. The new ICSPEA (Integrated CMA-SPEA) combines the powerful mutation concept of the CMA-ES with the evaluation scheme of the SPEA2. Tests on selected benchmark functions show the promising features of this new multi-objective optimisation algorithm.

Schlüsselwörter

CMA (Covariance Matrix Adaptation), SPEA2 (Strength Pareto Evolutionary Algorithm), Multi-objective Evolutionary Algorithms

Veröffentlichung

In: Reihe CI 222/07. SFB 531, Universität Dortmund, 2007

PDF-Anhang