An affinity propagation clustering based particle swarm optimizer for dynamic optimization

Multipopulation methods, which can enhance the population diversity, are well suited for dynamic optimization. However, there are still some challenges need to be tackled when multipopulation methods are employed, namely, how to avoid sensitive parameters when creating sub-populations, and how to effectively adapt to the changing optima continuously during the search process. Therefore, a novel multipopulation algorithm based on the affinity propagation clustering is proposed to address the above challenges. In the proposed method, affinity propagation clustering is applied for automatically creating sub-populations by message-passing process, which can avoid some extra parameters. Moreover, a simple but effective strategy, denoted as optimal particles relocation, is proposed for responding to environmental changes. In this strategy, the best particles in each sub-population are first stored in a memory. Then, local search is applied for helping the memory to quickly locate new peaks, if the environmental change has occurred. To validate the performance of the proposed algorithm, a variety of experiments have been conducted. The experimental results have demonstrated that the proposed algorithm performs robustly and competitively under different environments. (C) 2020 Elsevier B.V. All rights reserved.