Improved particle swarm optimization algorithm based on grouping and its application in hyperparameter optimization

In this article, an Improved Particle Swarm Optimization (IPSO) is proposed for solving global optimization and hyperparameter optimization. This improvement is proposed to reduce the probability of particles falling into local optimum and alleviate premature convergence and the imbalance between the exploitation and exploration of the Particle Swarm Optimization (PSO). The IPSO benefits from a new search policy named group-based update policy. The initial population of IPSO is grouped by the k-means to form a multisubpopulation, which increases the intragroup learning mechanism of particles and effectively enhances the balance between the exploitation and exploration. The performance of IPSO is evaluated on six representative test functions and one engineering problem. In all experiments, IPSO is compared with PSO and one other state-of-the-art metaheuristics. The results are also analyzed qualitatively and quantitatively. The experimental results show that IPSO is very competitive and often better than other algorithms in the experiments. The results of IPSO on the hyperparameter optimization problem demonstrate its efficiency and robustness.