Smart Grid Load Forecasting Using Hybrid Statistical and Deep Learning Models
Keywords:
Smart Grid; Load Forecasting; Hybrid Statistical Models; Deep Learning; Energy Demand Prediction.Abstract
This study proposes a hybrid statistical and deep learning framework for accurate electrical load forecasting in smart grid environments, focusing on improving demand prediction, grid stability, and energy management efficiency. The proposed methodology integrates historical electricity consumption records, meteorological variables, calendar information, and distributed energy generation data to characterize temporal variations in power demand. Statistical forecasting techniques, including Seasonal Autoregressive Integrated Moving Average, are combined with Long Short-Term Memory networks to capture linear trends, seasonal patterns, and nonlinear dependencies in electricity consumption. An adaptive model fusion strategy integrates individual predictions to improve forecasting robustness under changing demand conditions. Data preprocessing, feature engineering, and hyperparameter optimization are incorporated to enhance predictive performance. Model evaluation considers mean absolute error, root mean square error, mean absolute percentage error, and forecasting stability across different prediction horizons. The proposed framework supports intelligent demand response, efficient generation scheduling, and reliable operational planning in modern smart grid systems.