Optimal population size of particle swarm optimization for photovoltaic systems under partial shading condition

Abstract

Particle swarm optimization (PSO) is the most widely used soft computing algorithm in photovoltaic systems to address partial shading conditions (PSC). The success of the PSO run heavily depends on the initial population size (NP). A higher NP increases the probability of a global peak (GP) solution, but at the expense of a longer convergence time. To find the optimal value of NP, a trade-off is typically made between a high success rate and a reasonable convergence time. The most used trade-off method is a trial-and-error approach that lacks explicit guidelines and empirical evidence from detailed analysis, which can affect data reproducibility when different systems are used. Hence, this study proposes an empirical trade-off method based on the performance index (PI) indicator, which takes into account the weighted importance of success rate and convergence time. Furthermore, the impact of NP on achieving a successful PSO was empirically investigated, with the PSO tested with 16 different NPs ranging from 3 to 50, and 10,000 independent runs on various PSC problems. Overall, this study found that the best NP to use was 25, which had the best average PI value of 0.9373 for solving all PSC problems under consideration.