One day-ahead forecasting of mean hourly global solar irradiation for energy management systems purposes using artificial neural network modeling

Authors: K. P. Moustris, K. A. Kavvadias, A. I. Kokkosis


During the last decades, renewable energy sources (RES) have been established as one of the main solutions in coping with the future energy needs, without the negative effects caused by the use of fossil fuels. The exploitation of RES however, faces serious difficulties concerning the penetration limits in the electrical grid due to its stochastic and variable availability.One of the main parameters affecting the reliability of the RES system, compared to the local conventional power station, is the ability to forecast the RES availability for a few hours ahead.

The main goal of this work is the forecasting of global solar irradiation (GSI) on an horizontal plane, 24hours ahead, based only on historical meteorological data and artificial neural networks (ANN) modelling techniques. For that purpose, appropriate meteorological data have been recorded on minute intervals by a meteorological mast installed in Tilos Island, Greece from 17/03/2015 up to 20/12/2015.

According to the forecasting results, the coefficient of determination ranges between 0.500 and 0.851 as well as the root mean square error ranges between 0.065kWh/m2 and 0.105kWh/m2. Finally, the proposed forecasting ANN model shows a fairly good forecasting ability which is crucial for a better management of solar energy systems.

Published in: Power Generation, Transmission, Distribution and Energy Conversion (MedPower 2016), Mediterranean Conference on.
Key words: Solar Irradiation, forecasting, Artificial Neural Networks
Type: Article
Publisher: IET