Transparent artificial intelligence models using sequential minimal optimization and extreme Gradient Boosting for accurate performance prediction of solar farms

Solar energy is an important renewable source for lowering carbon emissions; forecasting solar power generation accurately to operate effectively is thus required. However, it is still challenging to estimate the power generation (PG) and performance ratio (PR) since wind speed (WS), monthly tilted irradiation (MTI), and air temperature (AT) are highly variable. Therefore, this current work employed interpretable machine learning approaches to handle these complex issues. Indeed, Extreme Gradient Boosting (XGBoost) and Sequential Minimal Optimization (SMO) were implemented to forecast PG and PR. Moreover, statistical measures like Coefficient of Determination (R 2 ), mean squared error, Kling-Gupta efficiency, and mean absolute percentage error were employed to see how well the model worked. Resultantly, the XGBoost model was better at predicting PR, with R 2 values of 0.9983 (train) and 0.8872 (test), compared to the SMO model's values of 0.9921 and 0.7484. Alternatively, radar plots demonstrated that XGBoost was better than SMO in terms of generalization and robustness. SHAP-based explainability was used to make the black models transparent, revealing that MTI had the most effect on PG and PR predictions, followed by WS and AT. Generally, explainable ML combined with SHAP offered high accuracy in forecasting solar farm output and enhancing energy management. • Interpretable ML models were developed to predict the performance of solar farms. • XGBoost outperformed SMO, achieving an R 2 of 0.9983 for the PG model. • SHAP analysis revealed MTI as the most influential factor, followed by WS and AT. • Radar plots confirmed XGBoost's superior generalization and robustness over SMO. • Combining SHAP with ML enhanced accuracy in solar energy forecasting.