An Interpretable Machine Learning Model for Daily Global Solar Radiation Prediction

Machine learning (ML) models are commonly used in solar modeling due to their high predictive accuracy. However, the predictions of these difficult explain and trust. This paper aims demonstrate utility two interpretation techniques improve ML models. We compared first performance Light Gradient Boosting (LightGBM) with three benchmark models, including multilayer perceptron (MLP), multiple linear regression (MLR), support-vector (SVR), for estimating global radiation (H) city Fez, Morocco. Then, most accurate model were explained by model-agnostic explanation techniques: permutation feature importance (PFI) Shapley additive explanations (SHAP). The results indicated that LightGBM (R2 = 0.9377, RMSE 0.4827 kWh/m2, MAE 0.3614 kWh/m2) provides similar accuracy as SVR, outperformed MLP MLR testing stage. Both PFI SHAP methods showed extraterrestrial (H0) sunshine duration fraction (SF) important parameters affect H estimation. Moreover, method established how each influences estimations. was further improved slightly after re-examination features, where combining H0, SF, RH better than all features.