Extracting accurate parameters of photovoltaic cell models via elite learning adaptive differential evolution

Photovoltaic power generation is becoming increasingly vital as the global call for environmental protection rises. Establishing an equivalent model for a photovoltaic cell and extracting accurate parameters of the model have a crucial role in supporting the fault diagnosis, performance analysis, and maximum power point tracking of the photovoltaic system. To better tackle this problem, an improved algorithm, i.e., Elite Learning Adaptive Differential Evolution (ELADE) is proposed. Four strategies, including the parameters adaptive strategy, elite learning strategy, chaotic last-place elimination strategy, and population size reduction strategy are combined to boost the exploitation process of differential evolution to effectively balance the ability to avoid local optimum and accelerate convergence speed. The suggested ELADE is applied to five photovoltaic cell models. Experimental results show that the maximum population size affects the performance of ELADE, and a recommended value 50 can promote it to achieve the most accurate and reliable parameters in comparison with other peer algorithms. Its superiority is further confirmed by two statistical test methods, including the Friedman for mean root mean square error (RMSE) values and Wilcoxon's rank-sum of RMSE values for each independent run. Besides, the influence of different strategies on ELADE is also empirically investigated, showing that the parameters adaptive strategy contributes the most.