Online 3-h forecasting of the power output from a BIPV system using satellite observations and ANN

Abstract

Photovoltaic (PV) systems are the reference technology in the solar-based electricity generation market. Rapid changes in solar radiation can alter PV power output; for this reason, knowledge of future atmospheric scenarios helps system operators to control the PV production in advance, reducing the instabilities that the electrical grid may suffer in electricity integration, and managing the auto consumption power output. With this is mind, we present a model to forecast (up to 3 h ahead) the building integrated photovoltaic (BIPV) system’s power output, which is installed on the roof of the Solar Energy Research Center (CIESOL), Almería, Spain. The satellite images have been combined with Artificial Neural Networks (ANN) primarily to predict power output using the lowest number of input variables. The results, which can be considered highly satisfactory, demonstrate the ANN’s prediction accuracy with an normalized root mean square error for all sky conditions of less than 26%, and with practically no deviation. We demonstrate how beneficial matching of two already proven techniques can bring about spectacular results in energy generation prediction for the BIPV system.