A New Selection Strategy for the Direction-based Multi-objective Evolutionary Algorithm

Abstract

The direction of improvement has been discussed and used to guide MOEAs during the search process towards the area of Pareto optimal set. One of typicalexamples using direction of improvement is the Directionbased Multi-objective Evolutionary Algorithm (DMEA).For DMEA, its authors introduced a novel algorithmincorporating the concept of direction of improvement.Our preliminary analysis showed that DMEA uses aselection procedure based on a weighted sum scheme for ahalf of the population. This is good for convergence, butit might make the population quickly losing diversity. Wepropose a new selection strategy to avoid this issue. Withthe new selection strategy, we make DMEA to be betterin balance between exploration and exploitation.To validate the performance of our proposed selectionstrategy for DMEA, we carried out a case study on a widerange of test problems and comparison with other MOEAs.We obtained quite good results on primary performancemetrics, namely the generation distance, inverse generationdistance and hypervolume. Our analysis on the resultsindicates the better performance of DMEA with the newselection strategy in comparison with the most popularMOEAs.