Building Power Demand and Energy Consumption Forecasting Using a Data-Driven Model: a Case Study in a Student Hostel

Abstract

Accurate forecasting of building power demand and energy consumption is essential for optimizing energy usage, improving efficiency, reducing costs, and ensuring sustainability. However, this prediction process is challenging due to factors such as variable occupancy, unpredictable occupant behavior, seasonal weather changes, data limitations, complex system interactions, and other external influences. This study develops a data-driven model based on historical electrical power data to predict the power demand and energy consumption of a student hostel. The historical data, recorded at five-minute intervals, was collected by logging the main incoming power supply using a power quality analyzer at the main switch block. Based on the power profile, the model was developed for four distinct time frames: falling, baseload, rising, and peak-load periods. Two key independent variables - minutes past midnight and type of day (weekday or weekend)—were considered as primary influences on power demand. Unlike previous models, this study employed MATLAB programming to optimize correlation modeling using the statistical approach of the power-law function. Results indicate that eighth- to ninth-degree polynomial fits provide the best power forecasting, achieving R² values as high as 0.9989. However, the prediction of power demand and energy consumption during peak-load periods on weekends was more complex, with a power correlation R² value of just 0.6100. Model accuracy assessments across different time frames and days showed that the developed model could predict power demand and energy consumption with a deviation of less than 5% compared to actual measurements. These findings demonstrate that a predictive model using only two independent variables, a power-law function, and polynomial fits up to the eighth and ninth degrees can effectively forecast power demand and energy consumption of the hostel. This model is expected to be valuable for future demand response (DR) programs, supporting the analysis of DR initiatives and the optimization of energy efficiency strategies. Future research could explore the integration of additional significant parameters alongside machine learning techniques to further enhance model accuracy. Factors such as outdoor air temperature, examination days, and a more detailed occupancy rate could be investigated and incorporated into future model development. This would allow for a more comprehensive evaluation of various energy consumption scenarios and their potential impact.