Distributed Photovoltaic Output Prediction Using Machine Learning and Numerical Model Combined Method

Abstract

Abstract In recent years, distributed photostatic (PV) has developed rapidly. Accurate prediction of distributed PV output can improve the reliable, efficient and economical grid-connected operation of distributed PV, and enhance the reliability of power supply. In this study, we first collected the datasets of distributed PV output and centralized PV station output in Dongan County, Yongzhou City, Hunan Province. Then, we analyzed the relationship between distributed PV station output and centralized PV output under different weather conditions, and constructed the centralized PV output prediction curve by combining machine learning and numerical prediction model, and finally obtained the distributed PV output prediction of the whole county. The results show that the correlation coefficient changes between the output of distributed PV stations and that of centralized PV stations under different weather scenarios (sunny, cloudy, etc.), but they all show a good consistency (the correlation coefficient of more than 0.6 accounts for more than 95% of the PV stations), indicating that the output change of centralized PV stations in one county can be used to characterize the output change of distributed stations. The distributed PV output prediction model built by LSTM and numerical model combined method has achieved a prediction accuracy of 90.4% for a single station and 94.31% for the whole county within 24 hours, both of which are 4%˜5% higher than the requirements of national standards. The model can be further extended to forecast the distributed PV output of the whole province.