EFFICIENCY INCREASE OF MULTI-OBJECTIVE GENETIC ALGORITHMS USING THE CLUSTERIZATION OF A POPULATION IN THE VARIABLE SPACE

Abstract

The paper introduces a new manner for improving of obtained by MOGA (Multi-Objective Genetic Algorithm) solutions. It is based on the concept of dividing the population into set of clusters according to solutions similarity. In different of most MOGA the clusterization of population is implemented in the variable space, enables to enhance diversity of population and to increase the number of non-dominated solutions. The special procedures for the clustering of current population and copying the clusters in the next population were developed. The dominance principal by fitness-value is used for clustering. The number of clusters depends on additional parameter the radius of cluster’s hypersphere that is determined experimentally. By the special rule the individuals corresponded to centroids of clusters are copied in the new population. The clusters are recalculated for every population. The influence of the radius cluster to the number of non-dominated solutions variation was studied. The cluster modification should be integrated into any multi-objective genetic algorithm. By the analytical evaluation has been studied, this MOGA modification has additional computationally complexity from linear to quadratic. In experiments it was tested with the evolutionary algorithms SPEA2, NSGA-II on the special benchmark problems (DTLZ) with a various number of criteria using the set of performance indices. The used clustering in the variable space algorithms were achieved a better distribution and convergence to the true Paretofront in some cases.