Ensemble strategy using particle swarm optimisation variant and enhanced local search capability

Abstract

Particle swarm optimisation is a population-based algorithm for evolutionary computation. A notable recent research direction has been to combine different effective mechanisms to enhance both exploration and exploitation capabilities while employing suitable mechanisms at appropriate instances in the evolutionary process. This study entailed the development of an ensemble strategy that uses a variant of the modified particle swarm optimisation algorithm with a covariance matrix adapted to the retreat phase and sequential quadratic programming. The modified particle swarm optimisation algorithm employs nonlinear population size reduction and uses the candidate elite best solution as the stochastic learning strategy and the fitness-distance balance as the terminal updating mechanism. The proposed algorithm was compared with the most recently proposed particle swarm optimisation-based variants through testing on CEC2017 benchmark functions. In the experimental results, the proposed method achieved the best ranking and exhibited excellent performance. Further effectiveness tests demonstrated that the proposed combination of algorithms and mechanisms exhibits tacit cooperation and significantly improves performance.