Photovoltaic Power Prediction from Medium-Range Weather Forecasts: a Real Case Study

The aim of this work is to utilize weather forecasts with a lead time from 6 h to 30 h as input data of a photovoltaic (PV) model to predict the AC power production. In order to always use the last forecasts, the inputs are updated every time there are new data, e.g., every 6 h. The ability of the model is tested on a residential PV plant for which global irradiance and electrical power are measured. The typical indicators of forecast accuracy in the PV applications are used: mean bias error and mean absolute error for both irradiance and power. However, they are normalized with respect to the standard irradiance and the PV rated power. Their values are generally adequate in clear sky and overcast conditions, remaining around the 10% limit.