Improved particle swarm optimizer for problems with variable evaluation time: Application to asymptotic homogenization

Abstract

AbstractThis work presents a new particle swarm optimizer (PSO)‐based metaheuristic algorithm designed to reduce overall computational cost without compromising optimization's precision in functions with variable evaluation time. The algorithm exploits the evaluation time gradient in addition to the convergence gradient attempting to reach the same convergence precision following a more economical path. The particle's newly incorporated time information is usually in contradiction to the past memories of best function evaluations thus degrading convergence. A technique is proposed in order to modulate the new cognitive input that consists of progressive reducing of its weight in order to confer the algorithm the appropriate time‐convergence balance. Results show that the proposed algorithm not only provides computational economy, but also unexpectedly improves convergence per se due to a better exploration in the initial stages of optimization. Its application in asymptotic homogenization of a cracked poroelastic medium confirms its superior performance compared to a series of alternative optimization algorithms. The proposed algorithm improvement allows to extend the applicability of PSO and PSO‐based algorithms to problems that were previously thought to be too computationally expensive for population‐based approaches.