Abstract
Guiding exemplar selection plays a crucial role in assisting particle swarm optimization (PSO) to gain satisfactory performance. To improve the effectiveness in helping PSO solve complex optimization problems with high effectiveness and efficiency deteriorates due to serious diversity loss, this paper devises a random shared local dominator guided scheme (RSLDG) for PSO, leading to a simple yet effective PSO variant named RSLDG-PSO. In contrast to existing studies, where each particle can only follow the guidance of the best position within its local area, RSLDG-PSO first randomly partitions the whole swarm into several sub-swarms and then identifies the best position of each sub-swarm. Then, all these local best positions are collected together to form a shared pool for all particles to learn. Subsequently, for each particle, a random local best position is chosen stochastically from the pool, along with its own historical experience, to guide its learning. In this way, highly diverse yet considerably promising exemplars are provided to update the swarm. Furthermore, to alleviate the sensitivity of RSLDG-PSO to parameters, this paper first devises an adaptive adjustment strategy for the sub-swarm size, and a dynamic strategy for adjusting the two coefficients. With the above schemes, RSLDG-PSO expectedly maintains a good dynamic balance between search diversity and search convergence to traverse complex solution space.