Forecasting Hot Water Consumption Demand for Residential Dwellings Using Hybrid Regression Technique

Residential buildings contribute approximately 30% of global greenhouse gas emissions, with electric water boilers accounting for around 20% of total energy consumption. The development of a smart renewable energy system and control requires adequate and accurate forecasting models to mitigate both short-term and long-term energy demand. Accurate forecasting of hot water consumption is crucial for optimizing load demand scheduling and grid energy manage-ment in residential dwellings. This study develops hybrid fore-casting models by combining a seasonal auto-regressive model (SARIMA) with machine learning models, specifically Support Vector Regression (SVR) and Artificial Neural Network (ANN). The study uses six months of hot water consumption data from residential electric boilers, categorized into hourly and daily scenarios. The models are evaluated based on performance metrics, including Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). Results obtain, indicate that the SARIMA-ANN model outperforms the SARIMA-SVR model in terms of both predictive accuracy and computational efficiency. The SARIMA-ANN model's superior feature mapping, in both long- and short-time frames, makes it optimal and robust for Model Predictive Control (MPC) applications and real-time use, especially in renewable energy grid integration. It offers a more scalable solution compared to the hyperplane-dependent SARIMA-SVR model.