Hybrid LSTM-XGBOOST Model for Sector-Specific Electricity Consumption Prediction in Iran: Incorporating Climate Scenarios and Sophisticated Machine Learning Methods

Abstract

Abstract: Precise forecasting of electricity use is crucial for efficient energy management, particularly in areas with diverse meteorological and economic circumstances. This study presents an innovative hybrid forecasting model that integrates Long Short-Term Memory (LSTM) networks with Extreme Gradient Boosting (XGBOOST) to anticipate power consumption in five principal sectors in Iran: industrial, agricultural, commercial, public, and residential. The model utilizes sophisticated feature selection and hyperparameter optimization to identify both linear and nonlinear consumption patterns, while integrating climate change scenarios (A1B, A1FI, and A1T) to evaluate future energy demand under diverse environmental conditions. The hybrid LSTM-XGBOOST model consistently outperforms individual models, exhibiting the lowest Mean Absolute Percentage Error (MAPE) values (4.20% to 10.79%) and the highest R² values across all sectors. The model's outstanding performance is particularly evident in its capacity to discern complex consumption patterns during peak periods and seasonal fluctuations. The study highlights the significant influence of regional characteristics, as evidenced by the exceptional forecast accuracy in provinces such as Bushehr and Semnan. It offers valuable insights for policymakers and energy system operators in Iran during their transition to renewable energy sources by proposing a robust and adaptable forecasting model that addresses sector-specific and regional issues, thereby advancing energy planning.