Enhancing A Multiobjective Evolutionary Algorithm Through Flexible Evolution

Abstract

In this paper the use of a powerful single-objective optimization methodology in Multi-objective Optimization Algorithms (MOEAs) is introduced. The Flexible Evolution concepts (FE) have been recently developed and proved its efficiency gains compared with several Evolutionary Algorithms solving single-objective challenging problems. The main feature of such concepts is the flexibility to self-adapt the internal behaviour of the algorithm to optimize its search capacity. In this paper we present the first attempt to incorporate FE into MOEAs. A real coded NSGA-II algorithm was modified replacing the crossover and mutation operators with the Sampling Engine of FE. Other two FE characteristics were implemented too: The Probabilistic Control Mechanism and the Enlarged Individual’s Code. The performance of the resulting algorithm has been compared with the classical NSGA-II using several test functions. The results obtained and presented show that FE_based algorithms have advantages over the classical ones, especially when optimizing highly multimodal complex functions.