Particle Swarm Optimization Enhanced with Kernel Principal Component Analysis

Particle swarm optimization (PSO) converges quickly in the initial stage of the search, and is essentially a random search algorithm. Such random search will inevitably lead to a premature convergence problem. In this study, we propose a novel particle swarm optimization enhanced by means of kernel principal component analysis (KPSO). The idea comes from particle swarm optimization imitates human social behavior. By introducing human social behavior, the optimal solution is searched from the overall driving swarm instead of considering only a single optimal particle, preventing particles premature. KPSO is tested on low-dimensional and high-dimensional benchmark functions. Experimental results show that compared with other PSO variants, the KPSO algorithm exhibits competitive performance in terms of accuracy and convergence speed, especially on high-dimensional problems. The KPSO algorithm is also applied to multi-fuel economic dispatch, and the results prove the effectiveness of the proposed method.