The Initialization of Evolutionary Multi-objective Optimization Algorithms

Abstract

Evolutionary algorithms are the most widely used meta-heuristics for solving multi objective optimization problems, and since all of these algorithms are population based, such as NSGAII, there are a set of factors that affect the final outcomes of these algorithms such as selection criteria, crossover, mutation and fitness evaluation. Unfortunately, little research sheds light at how to generate the initial population. The common method is to generate the initial population randomly. In this work, a set of initialization methods were examined such as, Latin hypercube sampling (LHS), Quasi-Random sampling and stratified sampling. Nonetheless. We also propose a modified version of Latin Hypercube sampling method called (Quasi_LHS) that uses Quasi random numbers as a backbone in its body. Furthermore, we propose a modified version of Stratified sampling method that uses Quasi-Random numbers to represent the intervals. For our research, a set of well known multi objective optimization problems were used in order to evaluate our initial population strategies using NSGAII algorithm. The results show that the proposed initialization methods (Quasi_LHS) and Quasi-based Stratified improved to some extent the quality of final results of the experiments.