Towards Sustainable and Transparent Solar Energy Systems using Explainable AI

The increase in the demand for clean and sustainable energy has enabled that new photovoltaic systems are being developed. Nevertheless, all of those methods still face the challenges of accurate prediction, system optimization, and interpret-ability in different environmental circumstances. This paper introduces an interpret-able AI framework by combining machine learning and explainability strategies to improve the sustainability, transparency, and performance of solar energy systems. The architecture includes four layers: data acquisition & pre-processing layer, AI modeling layer, explainability layer, and sustainability optimization layer. Several AI models such as XGBoost, Gradient Boosting, Random Forest and LSTM were compared by the real NSRDB data. The XGBoost showed better prediction performance (MAE = 16.88 kW, R2 =0.951, Pearson correlation = 0.978). Explanations were obtained from SHAP and LIME, their interpretations found solar irradiance and temperature to be significant features. In addition, sustainability indicators, including an energy efficiency improvement (12.6%), carbon offset (945 kg/y), battery usage improvement (18.4%) and maintenance downtime reduction (26.9%) validated that the framework’s effect can be felt across the actual spectrum. The interpretable AI model proposed here is a powerful tool not only for robust and accurate prediction, but also for actionable understanding in real-world energy planning, management, and policy design.