科研之友
微信
新浪微博
Facebook
分享链接
摘要
Particle swarm optimization (PSO) is a heuristics method based on a homogeneous population with identical search behavior. The simple homogeneous search is not always optimal. Recently, heterogeneous PSO (HPSO) search mechanisms have been developed to allow each particle to have different update rules for its position and velocity. However, in existing HPSOs, there are no supervisors to guide the particles to execute different search tasks. Differently, this paper investigates a new HPSO named SuPSO with a supervisor to allocate different tasks for each particle. In SuPSO the particles are separated into three groups by support vector machine (SVM): exploitation particles (EPs), diverse particles (DPs), and support particles (SPs). The supervisor enables the particle swarm to perform different tasks consisting of local search by EPs, global search by DPs, and linkage by SPs. The proposed SuPSO also employs a swarm migration mechanism, which, together with the DPs, enhances multi-modal optimization. The SuPSO algorithm was compared with other ten algorithms over twenty-eight benchmark functions (CEC2013). Based on the statistical analysis, we showed that SuPSO is statistically different from other ten algorithms with p-value less than 8x10(-3) and able to achieve good performance.
