Dynamic reference vectors and biased crossover use for inverse model based evolutionary multi-objective optimization with irregular Pareto fronts

Abstract

The recently developed inverse model based multi-objective evolutionary algorithm (IM-MOEA) has been shown to be an effective methodology for solving multi-objective optimization problems (MOPs) with regular Pareto fronts (PFs). Due to the limitations of uniformly distributed reference vectors and inverse model sampling, however, the IM-MOEA is challenged when solving MOPs with irregular PFs. To alleviate these limitations, both an external elitist archive and a biased crossover operator are integrated into the IM-MOEA. The primary role of the former is to dynamically adjust the reference vectors, which encourages the IM-MOEA to explore the sparse regions. The latter is used to improve the search efficiency of the IM-MOEA. By incorporating both features into the IM-MOEA, an enhanced IM-MOEA variant (E-IM-MOEA) is presented in this paper. Experimental studies were conducted on eighteen MOPs with irregular PFs to compare the E-IM-MOEA with six state-of-the-art multi-objective evolutionary algorithms. The experimental results show that the proposed approach has better overall performance.