A self-adaptive dynamic particle swarm optimizer

Abstract

A self-adaptive dynamic multi-swarm particle swarm optimizer (sDMS-PSO) is proposed. In PSO, three parameters should be given experimentally or empirically. While in the sDMS-PSO a self-adaptive strategy of parameters is embedded. One or more parameters are assigned to different swarms adaptively. In a single swarm, through specified iterations, the parameters achieving the maximum number of renewal of the local best solutions are recorded. Then the information of competitive arguments is shared among all of the swarms through generating new parameters using the saved part. Multiple swarms detect the arguments in various groups in parallel during the evolutionary process which accelerates the learning speed. What's more, sharing the information of the best parameters leads to faster convergence. A local search method of the quasi-Newton is included to enhance the ability of exploitation. The sDMS-PSO is tested on the set of benchmark functions provided by CEC2015. The results of the experiment are showed in the paper.