Comparative study of power forecasting methods for PV stations

In this paper, two power forecasting methods for PV systems, physical method and statistical method, are studied. A physical model based on the construction of PV systems and a NN statistical model based on historical data are set up. The impacts on forecasting accuracy of input data, such as solar irradiance, air temperature, cloud, humidity and sun position, for these two models are presented. Best input data models are founded for these two methods. Finally, the comparison of performance of the two forecasting models is investigated by a case study of a 1MW PV station. The nRMSE over one month of these two models are very close, i.e. around 10%-13%. The impacts of seasons on forecasting performance are presented. Moreover, by comparison, we found that the main origin of forecasting errors comes from the accuracy of weather prediction information, NWP. Thus, future improvement of power forecasting methods mainly relies on improvement of weather forecast in short-term forecasting application. Furthermore, real-time measured irradiance data can be considered to modify the model input to further improve super-short-term forecasting performance.