Machine learning approach for solar irradiance estimation on tilted surfaces in comparison with sky models prediction

Abstract

In this study, two supervised machine learning models (Extreme Gradient Boosting and K-nearest Neighbour) and four isotropic sky models (Liu and Jordan, Badescu, Koronakis, and Tian) were employed to estimate global solar radiation on daily data measured for one year period at the National Center for Energy, Research and Development (NCERD) at the University of Nigeria, Nsukka. Two solarimeters were employed to measure solar radiation: one measured solar radiation on a tilted surface at a 15° angle of tilt, facing south, and the other measured global horizontal solar radiation. The measured global horizontal solar radiation and the time and day number were used as input for the prediction process. Python computational software was used for model prediction, and the performance of each model was assessed using statistical methods such as mean bias error (MBE), mean absolute error (MAE), and root mean square error (RMSE) (RMSE). Compared to the measured data, it was discovered that the Extreme Gradient Boosting (XGBoost) algorithm offered the best performance with the least inaccuracy to sky models.