Explainable machine learning framework for near real-time energy forecasting and predictive control in smart manufacturing

Energy-intensive manufacturing systems face increasing pressure to improve energy efficiency while maintaining stable and responsive production operations. Although machine-learning models have shown strong capability in industrial forecasting applications, many existing approaches operate as black-box systems that provide limited transparency for operational decision-making. This lack of interpretability often reduces trust and limits practical deployment in live industrial environments. This study proposes an explainable Predictive Decision Support framework for near real-time energy forecasting in smart manufacturing systems. The framework integrates Extreme Gradient Boosting (XGBoost) with SHapley Additive exPlanations (SHAP) to combine predictive accuracy with interpretable operational insight. High-frequency industrial sensor data collected from a live steel production environment were used to evaluate the framework under realistic manufacturing conditions. The analysis incorporated operational variables including furnace temperature, operating pressure, coke flow rate, iron ore feed rate, and gas concentration measurements. The empirical results demonstrated that the XGBoost model achieved reliable forecasting performance with a Mean Absolute Error (MAE) of ۱.۲۷ MW, a Root Mean Square Error (RMSE) of ۱.۶۴ MW, and an R² value of ۰.۸۳. Comparative benchmarking against Support Vector Regression (SVR), Random Forest (RF), and Artificial Neural Networks (ANN) confirmed the robustness and computational suitability of the proposed approach. SHAP-based analysis identified furnace temperature and coke flow rate as the dominant contributors to energy consumption behavior, providing transparent explanations of prediction outcomes and supporting more informed operational decisions. The proposed framework demonstrates the practical feasibility of integrating explainable machine learning into industrial energy-management systems. It contributes to the development of transparent AI-driven decision-support architectures for Industry ۴.۰ environments.

کلیدواژه ها:

predictive control ، Real-Time Energy Forecasting ، Explainable AI ، Xgboost ، SHAP ، Industrial IoT