Forecasting short-term solar irradiance based on artificial neural networks and data from neighboring meteorological stations

• An intra-day solar irradiance forecasting method using neural networks (ANNs) from sensors.
• The influence of neighboring sensors in the short-term prediction is confirmed.
• The proposed ANNs improve error rates (<20% nRMSE) in nowcasting the solar irradiance.
• Validated ANNs models for 1–6 h time frames with sensors within a 55 km radius.

The study focused on the use of Artificial Neural Networks (ANN) in short-term prediction of Global Solar Irradiance (GSI). It introduces a new methodology based on observations made in parallel by neighboring sensors and values for different variables (temperature, humidity, pressure, wind and other estimates), using up to 900 inputs (higher dimensions). Experiments were carried out using ANN with different architectures and parameters in order to determine which of these generated the best GSI predictions for the various time frames studied (between 1 and 6 h). The results of the study allowed us to generate ANN models that predict short-term GSI with error rates less than 20% nRMSE. In addition, using observations from neighboring stations within a 55 km as a reference radius reduced error rates in predictions for time frames between 1 and 3 h, while the best predictions for time frames between 4 and 6 h were generated by ANNs that used only initial data from the station for which the prediction was being made.