Dimension selection: an innovative metaheuristic strategy for particle swarm optimization

Abstract Particle swarm optimization (PSO) is a prominent metaheuristic algorithm that has demonstrated remarkable efficiency in tackling diverse optimization problems. Nonetheless, the conventional PSO algorithm and its state-of-the-art variants update the global best position entirely although better results can be achieved if only certain dimensions are updated. This sub-optimal replacement can limit the potential of PSO to obtain better optimization performance. To tackle this issue, this work proposes a new strategy called dimension selection that aims to decide which dimensions should be updated if a better solution is found. Dimension selection is a binary problem where a value of 1 indicates that a dimension of the global best position should be updated while a value of 0 means that a dimension should keep its value. The dimension selection strategy is integrated with PSO resulting in a new algorithm called dimension selection PSO (DSPSO). To solve the dimension selection problem, seven well-known and recent binary algorithms are implemented. To test the effectiveness of DSPSO, comprehensive experiments are conducted using two of the most complex and challenging test suites: CEC2017 and CEC2020. Moreover, DSPSO is tested on four constrained engineering problems. The performance of DSPSO is compared with seven well-established and high-performance PSO and non-PSO algorithms. The Wilcoxon rank-sum and Friedman statistical tests show that DSPSO significantly outperforms other competing algorithms on the majority of the considered problems. The promising results of DSPSO motivate other researchers to apply the dimension selection approach to enhance the performance of existing or new metaheuristic algorithms. The source codes of the DSPSO algorithm are publicly available at https://nimakhodadadi.com/algorithms-%2B-codes .