Deep learning and ensemble learning models for individual appliance level short-term load forecasting in buildings

Accurate forecasting of electrical loads is a critical task for optimal control strategies and building energy management. Thanks to technological advances of the internet of things and the storage of energy data, using machine learning and deep learning models for predicting the power consumption of buildings attracts great attention from researchers. Most previous studies mainly focused on the aggregate level load forecasting of the whole building. This paper presents a novel investigation that aims to predict the short-term building consumption loads at the individual appliance level. Our study focuses on comparing the forecasting performance of ensemble-learning models (Random Forest and XGBoost) based on bagging and boosting techniques, and recurrent deep learning models (LSTM and GRU) for an hour-ahead and a day-ahead forecasting horizon. The models are trained and tested on a real dataset that is collected from a residential household in Vietnam for the assessment.