Adequate is better: particle swarm optimization with limited-information

摘要

Based on the interaction of individuals, particle swarm optimization (PSO) is a well-recognized algorithm to find optima in search space. In its canonical version, the trajectory of each particle is usually influenced by the best performer among its neighborhood, which thus ignores some useful information from other neighbors. To capture information of all the neighbors, the fully informed PSO is proposed, which, however, may bring redundant information into the search process. Motivated by both scenarios, here we present a particle swarm optimization with limited information, which provides each particle adequate information yet avoids the waste of information. By means of systematic analysis for the widely-used standard test functions, it is unveiled that our new algorithm outperforms both canonical PSO and fully informed PSO, especially for multimodal test functions. We further investigate the underlying mechanism from a microscopic point of view, revealing that moderate velocity, moderate diversity and best motion consensus facilitate a good balance between exploration and exploitation, which results in the good performance.