Estimation of Daily Direct Normal Solar Irradiation Using Machine-Learning Methods

The sizing and simulation of all solar systems require the availability of reliable measurements of solar radiation at different time steps. Unfortunately, solar radiation measurements are not readily available for most worldwide locations. For this reason, it is desirable to develop accurate prediction models by developing relationships between available meteorological data and solar irradiation. Artificial Neural Networks (ANN) have been widely used for the estimation of different solar irradiation components. Recently, some machine learning methods have been reported and appear to be very promising. In this paper, we are interested in comparing the performance of ANN and three ensemble methods (Bagging, Boosting and Random Forests) in estimating the daily direct normal (DNI) solar irradiation from some commonly measured meteorological variables. Our study is performed using measurements data from five Moroccan cities: Oujda, Missour, Erfoud, Zagora, and Tan-Tan. The achieved results show that all developed models give good performances on training and validation datasets with a normalized Root Mean Squared Error (nRMSE) < 20%.