AI-based hourly electricity consumption prediction for households and residents

This research aims to develop an XGBoost-based predictive model for accurately forecasting hourly electricity consumption in diverse household scenarios. By leveraging household-specific details and time-specific data from the IDEAL dataset, which encompasses over 200 UK households, the study delves into precise electricity usage forecasting. Data pre-processing is employed to ensure data reliability and reduce the computational costs, while XGBoost's robust learning capability renders it particularly suitable for time series forecasting. Two training techniques, grid search and cross validation were introduced to find the most proper parameters, which assured the quality of the model. From the results, it is evident that the demand for electricity is primarily determined by temporal factors (hour, days, month, quarter, year, weekday) and marginally related to household features (number of residents, build era, income, install type). The developed model effectively captures the fundamental patterns of electricity demand, prompting both consumers and producers to embrace energy-efficient measures, paving the way for a more sustainable future. Meanwhile, energy experts could refer to the model with actual values to formulate an energy generating strategy based on the number of population and household for a specific location. This prediction can provide a dispatching basis for various smart grids.