A Combined Model for Ultra-Short-Term PV Forecasting Based on SOM Clustering

Abstract

A combined model for ultra-short-term PV forecasting based on SOM clustering is proposed to improve the accuracy of PV power prediction and reduce the impact of the randomness of PV power generation on the power system. At first, the key factors are first selected as inputs to the model by calculating the Pearson correlation coefficients between each factor and PV power. Second, to eliminate the influence of season on weather classification and the coupling relationship between many meteorological factors, the key factors are standardized and weighted summed month by month to obtain the classification index Sky Condition Factor (SCF). Then, the SCF is clustered unsupervised by self-organizing mapping (SOM) neural network, to classify the sample data into three weather types and construct CNN prediction models under different weather types respectively. The results show that the combined model proposed in this paper has obviously improve the accuracy of PV output power prediction for different weather conditions.