Sensitivity analysis of control parameters in particle swarm optimization

Abstract

Particle Swarm Optimization (PSO) is a powerful nature-inspired metaheuristic optimization method that may determine optimal solutions of engineering problems in fewer evaluations compared to other optimization methods. However, the literature shows that PSO may suffer from converging prematurely to a local solution, and this occurs due to poor tuning of the control parameters in PSO. In this paper, an extensive parametric sensitivity analysis was conducted to understand the impact of the individual control parameters and their respective influence on the performance of PSO. A benchmark constrained optimization problem was considered for studying PSO by modifying each parameter one-at-a-time. Therefore, initially, a constraint handling technique was formulated to allow particles to update their best historical solutions according to the feasibility. Results of the sensitivity analysis revealed that PSO was most sensitive to the inertia weight, cognitive component, and social component. The optimal parameter set, determined from the sensitivity analysis, was verified by comparison with metaheuristic methods. The verification study shows that the proposed parameter setting outperformed the other methods in all but one case, where it performed competently