Development and assessment of solar radiation forecasting models based on operational data

Abstract

Operational forecasting of solar radiation is critical for better decision-making by solar energy system operators, due to the variability of energy resources and demand. Although the numerical weather forecasting (NWP) model can predict solar radiation variables, there are often significant errors, especially in direct normal irradiation (DNI), which are influenced by the type and concentration of aerosols and clouds. This paper presents an artificial neural network (ANN) based method to generate operational DNI forecasts using weather and aerosol forecast data from the European Center for medium-range weather forecasts (ECMWF) and Copernicus atmospheric monitoring service (CAMS) respectively. The ANN model is designed to predict weather and aerosol variables at a certain time as input, while other models use the DNI forecast improvement period before the instant forecast. The model was developed using North Sumatra location observations and obtained DNI forecasting results every 10 minutes on the first day with DNI forecasting compared to the initial forecasting which was scaled down with the R2, mean absolute error (MAE), and relative mean square error (RMSE) models were 0.6753, 151.2, and 210.2 W/m2, so that and provides good agreement with experimental data.