Integrating particle swarm optimization with learning automata to solve optimization problems in noisy environment

Particle Swarm Optimization (PSO) is a brilliant evolutionary algorithm adaptable to various kinds of optimization problems in distinct fields. However, when facing noisy environments, its performance suffers from unexpected noise. To address this issue, one of the widely-used mechanisms is the resampling method that is based on the fact that the true objective value can be achieved by re-evaluations. Such method allocates a fixed number of re-evaluations before running but cannot change allocations according to the current environment adaptively. It may result in the waste of re-evaluations allocated to unpromising candidate particles. This paper proposes a novel hybrid approach by integrating PSO with Learning Automata (LAs) in noisy environments. LAs are well-known for their self-adaption, automatic learning capability as well as low computational complexity. They are able to converge in different situations. The proposed hybrid approach achieves much faster convergence than the existing ones by performing fewer re-evaluations in simple environments than in complex one automatically. This mechanism enables it to find the best particle efficiently. With its self-adaption and automatic learning capability, it leads to a more accurate and faster algorithm. Besides, its distinct selection mechanism helps it achieve a significantly lower computational complexity than that of the-state-of-the-art resampling methods. Through experiments on 20 large-scale benchmark functions subject to different levels of noise, it is validated that, the proposed approach is able to achieve much better performance results in terms of accuracy and convergence rate than the existing ones.