Photovoltaic power forecasting method based on adaptive classification strategy and HO-SVR algorithm

Abstract The quality of similar sample data determines the accuracy of photovoltaic (PV) power forecasting. However, under different time and space scales, the main meteorological characteristics affecting PV power and their mechanisms are different, which seriously affects the quality of similar samples. An adaptive classification strategy is proposed to filter historical similar samples. Firstly, path analysis (PA) adaptation is utilized to determine the main meteorological characteristics affecting PV power at different spatial and temporal scales, as well as the determining coefficient of each meteorological characteristic on PV power. Secondly, a negative feedback strategy based on the distribution factor and fitness function value of the forecasting model is claimed, which can adaptive adjust the selection time range of the historical similar samples until the forecasting model with higher fitting degree obtained based on the hybrid optimization support vector regression (HO-SVR) algorithm training. Finally, the validity and practicability of the forecasting model are verified by historical measured meteorological data and power data of a PV power plant.