An Explainable AI–Driven Type-2 Fuzzy Decision Support Framework for Sustainable Energy Management in Smart Grids

Abstract

The increasing integration of renewable energy resources, distributed energy systems, and intelligent monitoring technologies has transformed traditional power networks into complex smart-grid environments. Although Artificial Intelligence (AI) and Machine Learning (ML) techniques have significantly improved forecasting accuracy and operational efficiency, the lack of transparency in many AI models remains a major challenge for critical energy-management applications. In addition, the inherent uncertainty associated with renewable energy generation, electricity demand, and market conditions further complicates decision-making processes in smart grids. To address these challenges, this paper proposes an Explainable Artificial Intelligence (XAI)–Driven Type-2 Fuzzy Decision Support Framework for sustainable energy management in smart grids. The proposed framework integrates Extreme Gradient Boosting (XGBoost)-based forecasting, SHapley Additive exPlanations (SHAP), and an Interval Type-2 Fuzzy Inference System (IT2FIS) into a unified architecture. First, the forecasting module predicts future electricity demand and renewable energy generation using historical operational and environmental data. Subsequently, SHAP is employed to provide both global and local explanations by quantifying the contribution of input variables to forecasting outcomes. The extracted explainability information is then incorporated into the Type-2 fuzzy decision-support module, which generates interpretable and uncertainty-aware energy-management actions under dynamic operating conditions. The proposed framework aims to simultaneously improve prediction accuracy, enhance transparency, and increase the robustness of decision-making in the presence of uncertainty. Performance evaluation is designed using standard forecasting metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²), together with sustainability-oriented indicators such as Renewable Energy Utilization rate (REU), Peak Load Reduction (PLR), Energy Cost Saving Ratio (CSR), and Carbon Emission Reduction (CER). Compared with conventional ML and Type-1 fuzzy approaches, the proposed framework is expected to provide more reliable predictions, greater interpretability, and improved sustainability performance. The findings of this study contribute to the development of trustworthy and human-centered intelligent energy-management systems and provide a promising pathway toward transparent, resilient, and sustainable smart-grid operation.