Intra-hour PV Power Forecasting based on Multi-source Data and PSC-SVR Model

Accurate photovoltaic power forecasting (PVPF) can ensure the stability of grid operation and the reliability of economic dispatch. However, PV power generation is greatly affected by weather conditions, thus more external information needs to be combined to improve the forecasting performance. Based on the measured data and NWP data, this paper proposes a PSC-SVR model to forecast the intra-hour PV power. The model consists of PV physical model (PSC) and support vector regression (SVR). PSC converts the irradiance as a priori feature of the model, including calculating the sun position, irradiance separation and transposition model. SVR is used for the two-stage data fusion of NWP power modeling and power forecasting. This method was verified on 4 power stations in the public data set PVOD. When forecasting PV power in the next one hour, the average RMSE/MAE of all stations is 1.927 MW/1.167 MW, which decreases by 40%/55% compared with persistent model, 6%/7% compared with NWP correction forecasting method, 9%/13% and 32%/34% compared with the single measured data and single NWP data. The above results show the effectiveness of our proposed model in multi-source data fusion and PV power forecasting.