Improving Search Accuracy in Large-Scale Biased Multiobjective Optimization Through Local Search

Biased multiobjective optimization problems pose a challenge for evolutionary algorithms in obtaining high-accuracy solutions, and as the number of decision variables increases, this challenge becomes increasingly difficult to overcome. To address this issue, we propose a three-particle-based local search method (TPS) for multiobjective evolutionary algorithms (MOEAs). The main concept is to use three particles to maintain three equidistant values of a decision variable and gradually approach the local optimal value by adaptively adjusting their differences. Specifically, the TPS maintains a population with three particles and uses five proposed population state-transition operations to gradually move these three particles to a better state. A local optimal value can be obtained when these three particles become indistinguishable. The TPS is then embedded into an MOEA to form a new algorithm, called MOEA/TPS. To enable the TPS to search along the convergence and diversity directions, the two aggregation functions of the target problem are alternately used. Compared with twelve competitive MOEAs on various biased test problems with 30 to 2000 decision variables, our proposed algorithm demonstrates significant advantages in obtaining high-accuracy solutions.