Self-Adaptive Mechanism for Multi-objective Evolutionary Algorithms

Evolutionary algorithms can efficiently solve multi-objective optimization problems (MOPs) by obtaining diverse and near-optimal solution sets. However, the performance of multi-objective evolutionary algorithms (MOEAs) is often limited by the suitability of their corresponding parameter settings with respect to different optimization problems. The tuning of the parameters is a crucial task which concerns resolving the conflicting goals of convergence and diversity. Moreover, parameter tuning is a time-consuming trial-and-error optimization process which restricts the applicability of MOEAs to provide real-time decision support. To address this issue, we propose a self-adaptive mechanism (SAM) to exploit and optimize the balance between exploration and exploitation during the evolutionary search. This "explore first and exploit later" approach is addressed through the automated and dynamic adjustment of the distribution index of the simulated binary crossover (SBX) operator. Our experimental results suggest that SAM can produce satisfactory results for different problem sets without having to predefine/pre-optimize the MOEA's parameters. SAM can effectively alleviate the tedious process of parameter tuning thus making on-line decision support using MOEA more feasible.