Adaptive niching particle swarm optimization with local search for multimodal optimization

Abstract

Multimodal optimization problems (MMOPs), in which multiple optimal solutions need to be found for decision-makers, are common in real-world applications. Finding as many peaks as feasible and enhancing the accuracy of solutions on already identified peaks are the objectives of solving MMOPs. An adaptive particle swarm optimization (PSO) based on speciation, species regulation, and local search, termed SR-PSO-MAES, is proposed to achieve these objectives. First, an adaptive speciation strategy is introduced to divide the population, which forms species according to the similarity of particles and does not need to set the radius of species in advance. Second, a species regulation strategy is proposed to avoid a few species occupying most of the population during the iterative process. Third, a matrix adaptation evolution strategy (MAES) with a restart scheme as a local search strategy is used to evaluate the species for further improving the accuracy of the solutions. By comparing with 15 state-of-the-art multimodal optimization algorithms, the experimental results through a benchmark test problem and a real wet spinning process validate the superiority of the proposed SR-PSO-MAES algorithm.