A Hybrid Ensemble Machine Learning Framework for Short-Term Energy Demand Forecasting with Seasonal Consumption Analysis

Abstract

Electricity demand forecasting plays a vital role in energy planning, load balancing, cost reduction, and reliable operation of power distribution systems. However, short-term electricity consumption prediction remains challenging due to the nonlinear influence of temporal variations, weather conditions, consumer behavior, weekends, holidays, and seasonal demand fluctuations. This study proposes a hybrid ensemble machine learning framework for short-term electricity demand forecasting using hourly electricity consumption data of Kolhapur city obtained from the Maharashtra State Electricity Board for the period 2020-2024. The proposed framework integrates Linear Regression, Random Forest, and XGBoost through an optimized weighted ensemble strategy to capture both linear and nonlinear electricity demand patterns. In addition to forecasting, the study analyzes weekday–weekend variation, holiday-period demand behavior, monthly demand distribution, seasonal consumption trends, and daytime–nighttime load patterns. The model performance is evaluated using Root Mean Square Error, Mean Absolute Error, and coefficient of determination. Experimental results show that the proposed hybrid ensemble model achieves the best forecasting performance, with an approximate RMSE of 145, MAE of 95, and R2 score of 0.98, outperforming Linear Regression, Random Forest, and XGBoost models individually. The findings indicate that combining ensemble machine learning with temporal and seasonal consumption analysis can provide accurate and reliable short-term electricity demand forecasts for city-level energy management and smart-grid planning.