AN INTELLIGENT HYBRID LSTM–XGBOOST FRAMEWORK FOR TIME-SERIES ENERGY DEMAND FORECASTING IN SMART ELECTRICAL POWER SYSTEMS USING SHAP-BASED FEATURE OPTIMIZATION

Abstract

The rapid expansion of smart electrical infrastructures, renewable energy integration, and dynamic electricity consumption patterns has significantly increased the complexity of accurate energy demand forecasting in modern Electrical Engineering environments. Conventional forecasting approaches often suffer from limited predictive accuracy, weak adaptability to nonlinear temporal variations, and insufficient interpretability when processing large-scale time-series energy datasets. To overcome these limitations, this study proposes an intelligent hybrid forecasting framework based on Deep Learning and Machine Learning for efficient energy demand prediction in smart electrical power systems. The proposed framework combines Long Short-Term Memory (LSTM) networks to capture long-term temporal dependencies and sequential load behavior, while XGBoost is employed to enhance regression accuracy, predictive stability, and computational efficiency. In addition, Explainable Artificial Intelligence-based feature optimization is integrated to identify the most influential forecasting parameters and improve model transparency and interpretability. The proposed methodology includes data preprocessing, normalization, temporal feature extraction, SHAP-driven feature selection, hybrid model training, and comparative performance evaluation using multiple forecasting metrics, including accuracy, Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Experimental results demonstrate that the proposed hybrid LSTM–XGBoost framework achieves a forecasting accuracy of 98.2%, outperforming conventional machine learning models and standalone deep learning approaches. The model reduced RMSE by 21.4% and MAE by 18.7% compared with traditional forecasting techniques, while also improving prediction stability under dynamic load fluctuations and seasonal demand variations. Furthermore, SHAP-based analysis revealed that historical load demand, temperature variations, peak-hour consumption, and renewable energy penetration were among the most influential features affecting forecasting performance. The overall findings confirm that the proposed intelligent hybrid framework provides a reliable, scalable, and interpretable solution for real-time energy demand forecasting, smart grid optimization, and intelligent energy management applications in next-generation electrical power systems