Probability-optimal leader comprehensive learning particle swarm optimization with Bayesian iteration

Abstract

In this paper, a novel comprehensive learning particle swarm optimization algorithm, which is based on the Bayesian iteration method and named as Bayesian comprehensive learning particle swarm optimization (BCLPSO), is proposed. In the original PSO, the flying direction of each particle is based on its own historical best position and global optimum. This updating mechanism, however, easily falls into the local optimum, and the potential optimum solution may be ignored in the iteration and update process. Therefore, the BCLPSO is designed to facilitate discovering potential solution and avoid the problem of premature convergence. In the BCLPSO algorithm, the exemplar of the swarm is not the global best position but the particle location with the largest posterior probability based on the Bayesian formula. The posterior probability is developed by historical prior information. This means that the posterior probability can inherit the historical information of particles that may be exploited. In this way, the swarm diversity can be preserved to prevent premature convergence. The BCLPSO is experimentally validated on the CEC2017 benchmark functions and compared with other state-of-the-art particle swarm optimization algorithms. The results show that BCLPSO outperforms other comparative PSO variants on the CEC 2017 test suite. Furthermore, the algorithm is applied to the quality control process of an automated welding production line for the automobile body and is found to exhibit superior performance.