Comparative analysis of machine learning models for prediction and forecasting of electric water boilers energy consumption

• Evaluation of ARIMA, SVR, and ANN for predicting EWB energy use in buildings. • Examine short and long time-step’s impact on prediction accuracy. • ANN shows superior performance in capturing consumption patterns. • Balances accuracy vs. computational cost for prediction models. • Framework for EWBs forecasting, applicable to MPC, EVs, and smart grids. In the quest to accurately predict energy consumption in high-rise residential buildings for model predictive control implementation to enhance energy efficiency and address energy poverty. This study compares two supervised learning algorithms (support vector regression (SVR) and multi-layer perceptron (MLP) artificial neural network (ANN)) and a regression-based forecasting technique (ARIMA). These methods are used to predict and forecast energy consumption for electric water boilers in a residential building comprising 67 apartments, based on six months of data as a case study. The unique approach of the study involves five distinct scenarios with different time steps: 1) 10-minute intervals of daily energy consumption, 2) hourly daily consumption, 3) daily consumption during the winter season, 4) daily consumption during the spring season, and 5) daily energy consumption for six months. These timesteps are used for the prediction (one time-step ahead) and forecasting (many time-steps ahead) of electricity consumption by the hot water boilers. The objective is to evaluate the effectiveness, accuracy, and computational efficiency of these scenarios in both prediction and forecasting. Furthermore, the paper presents the performance metrics of each method, highlighting their respective strengths and weaknesses. In conclusion, the SVR model demonstrated average overall performance compared to the ANN and ARIMA models, with noteworthy results in scenario 2 with one-hour time steps. The paper offers recommendations and suggests areas for future research and potential model optimisations to enhance demand-side management and advance energy conservation initiatives in residential buildings.