Embedding a Repair Operator in Evolutionary Single and Multi-objective Algorithms - An Exploitation-Exploration Perspective.

Evolutionary algorithms (EAs) are population-based search and optimization methods whose efficacy strongly depends on a fine balance between exploitation caused mainly by its selection operators and exploration introduced mainly by its variation (crossover and mutation) operators. An attempt to improve an EA’s performance by simply adding a new and apparently promising operator may turn counter-productive, as it may trigger an imbalance between the exploitation-exploration trade-off. This necessitates a proper understanding of mechanisms to restore the balance while accommodating a new operator. In this paper, we introduce a new repair operator based on an AI-based mapping between past and current good population members to improve an EA’s convergence properties. This operator is applied to problems with different characteristics, including single-objective (with single and multiple global optima) and multi-objective problems. The focus in this paper is to highlight the importance of restoring the exploitation-exploration balance, when a new operator is introduced. We show how different combinations of problems and EA characteristics pose different opportunities for restoration of this balance, enabling the repair-based EAs/EMOs to outperform the original EAs/EMOs in most cases.