Accurate short-term electricity demand forecasting is essential for ensuring reliable power supply, optimizing grid operations, and supporting sustainable energy planning. In Tamil Nadu, seasonal variability, economic growth, and climatic anomalies make peak demand prediction particularly challenging. This study aims to develop a robust forecasting framework that addresses these challenges by integrating statistical time-series modeling with machine learning techniques. Historical monthly peak demand data from April 2006 to March 2023, sourced from the Tamil Nadu Transmission Corporation, were analyzed alongside climate and socio-economic variables. Preprocessing involved missing value imputation, stationarity checks, and feature engineering, followed by model development using Autoregressive Integrated Moving Average (ARIMA), Vector Autoregression (VAR), and a hybrid VAR–machine learning ensemble incorporating Lasso, Ridge, and XGBoost regressors. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). Results show that the ARIMA model achieved the lowest MAPE (1.26%), outperforming VAR and hybrid approaches, particularly in capturing seasonal trends. However, error margins increased during anomalous months influenced by extreme weather events, highlighting the need for incorporating additional real-time predictors. This research demonstrates that a well-calibrated ARIMA model offers a reliable and practical solution for Tamil Nadu’s short-term peak demand forecasting, providing actionable insights for policymakers, utility planners, and grid operators.
Suriya et al. (Wed,) studied this question.