On the prediction of daily global solar radiation using temperature as input. An application of hybrid machine learners to the six climatic Moroccan zones

Forecasting of solar radiation intensity is a necessity for the establishment of solar energy projects and for decision-making in other related fields. Current prediction models/methods are site-dependent and their performance/accuracy outside the area of application is debatable. Temperature-based solar radiation models are highly recommended in areas where only air temperature data is available. Therefore, the purpose of this study is to evaluate the prediction accuracy of 42 existing temperature-based solar radiation models in forecasting the daily global solar radiation (DGSR) on horizontal surface for the six climatic zones of Morocco. In the first time, the models were assessed using only the least square method. Then, four Machine Learners models (SVM, Decision Tree, Gaussian Regression and Linear Regression) were employed as optimizers to improve the accuracy prediction of the models. The results differ from model to another based on their values of MBE, MSE, RMSE, σ and R2. Two methods were employed for ordering the studied models: the Performance score and the Taylor diagram. Long term meteorological data was used in the evaluation processes. The correlation R2 of the optimized models changes from 0.80 to 0.95 for all skies and from 0.95 to 0.98 for clear skies.