Comparative Study of Different Inertia Weight Strategies in Particle Swarm Optimization Based on Actual Computational Time Cost

Abstract: The particle swarm optimization (PSO) algorithm is attracting a lot of research attention due to its superior performance over other swarm-based algorithms. However, one of the major challenges facing PSO is the tendency to fall into local optima, which is known as premature convergence. The inertia weight variable was introduced into PSO to solve this problem by balancing the relationship between exploration and exploitation stages in swarm activity within a given search space. Many studies have proposed different inertia weight strategies to improve on convergence performance of PSO including, Constant Inertia Weight (CIW), Linearly Decreasing Inertia Weight (LDIW), Exponential Inertia Weight (EIW), Chaotic Inertia Weight (CHIW), Nonlinear Decreasing Inertia Weight (NDIW), Adaptive Inertia Weight (AIW), Random Inertia Weight (RIW) and Time Varying Inertia Weight (TVIW). However, these strategies have also introduced varying levels of computational complexities into the PSO algorithm. This study compares eight different inertia weight strategies based on their computational time cost, in order to propose the most efficient strategy. The experiments were carried out using PSO implementation in a Cloudsim simulation environment based on actual computational runtime of each inertia weight strategy. In summary, the chaotic inertia weight strategy has the lowest average runtime of 3610552.27 microseconds, followed by TVIW = 3611035.51 LDIW = 3611035.95, CIW = 3611044.09, AIW = 3611539.87, NDIW = 3612029.75, RIW = 3612520.84, and EIW = 3612524.36.