Prediction of Solar Radiation Intensity using Extreme Learning Machine

Hadi Suyono, Hari Santoso, Rini Nur Hasanah, Unggul Wibawa, Ismail Musirin
2018 Indonesian Journal of Electrical Engineering and Computer Science  
The generated energy capacity at a solar power plant depends on the availability of solar radiation. In some regions, solar radiation is not always available throughout the day, or even week, depending on the weather and climate in the area. To be able to produce energy optimally throughout the year, the availability of solar radiation needs to be predicted based on the weather and climate behavior data. Many methods have been so far used to predict the availability of solar radiation, either
more » ... radiation, either by mathematical approach, statistical probability, or even artificial intelligence-based methods. This paper describes a method of predicting the availability of solar radiation using the Extreme Learning Machine (ELM) method. It is based on the artificial intelligence methods and known to have a good prediction accuracy. To measure the performance of the ELM method, a conventional forecasting method using the Multiple Linear Regression (MLR) method has been used as a comparison. The implementation of both the ELM and MLR methods has been tested using the solar radiation data of the Basel City, Switzerland, which are available to public. Five years of data have been divided into training data and testing data for 6 case-studies considered. Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) have been used as the parameters to measure the prediction results based on the actual data analysis. The results show that the obtained average values of RMSE and MAE by using the ELM method respectively are 122.45 W/m<sup>2</sup> and 84.04 W/m<sup>2</sup>, while using the MLR method they are 141.18 W/m<sup>2</sup> and 104.87 W/m<sup>2</sup> respectively. It means that the ELM method proved to perform better than the MLR method, giving 15.29% better value of RMSE parameter and 24.79% better value of MAE parameter.
doi:10.11591/ijeecs.v12.i2.pp691-698 fatcat:pcrbegqdijhg7i3mgctwkp6ida