Short-term photovoltaic generation forecasting based on similar day selection and extreme learning machine
Ping Luo, Shuncun Zhu, Lujie Han, Qiaoyong Chen · 2017
A new short-term PV generation forecast approach is proposed to improve the prediction accuracy. First, the fuzzy clustering method and gray correlation coefficient method is adopted to select the similar days. Then the PV output historical data of these similar days are used as training samples to train the extreme learning machine (ELM) model which has weights and thresholds optimized by Genetic algorithm(GA) to overcome the over fitting phenomena of ELM which caused by the randomly generation of weights and thresholds. At last the predicted results can be obtained by using the trained neural network and local meteorological data. Simulation experiments based on real data were compared with the traditional ELM and BP forecasting method. The simulation results show that the forecasting result based on similar day selection and GA-ELM network has the best prediction accuracy.