A Global Data-driven Forecasting Approach for Buildings Energy Demand Prediction

Energy demand prediction is a key factor for buildings operation optimization, and energy conservations. Recently, the rapid advancement of sensing technology, and smart meters in the buildings sector has allowed to record and collect large amount of building energy datasets which can provides the opportunity to understand how these buildings are being used to optimize and reduce their daily energy usage. Nevertheless, the majority of existing works have mostly been focused on the utilization of data-driven approaches to predict the future energy demand for one building at once. This study attempts to address this gap by proposing a Global Data-driven Forecasting Approach which has been trained with several time series datasets from a group of residential buildings. Utilizing the proposed approach offers numerous benefits including: (I) better generalisation ability on predicting the future energy demand for a new building that is coming from new dataset, and (II) facilitating the prediction even when the training set is limited. To assess the effectiveness of this approach, this work aims to compare the prediction from six different data-driven approaches by forecasting the daily buildings electricity consumption. The empirical analysis showed that the proposed global forecasting approach is an effective and promising framework for building energy predictions.