Deep Learning-based Predictive Modeling of Building Energy Usage

Buildings are responsible for around 40% of the total global energy consumption and the increase in the energy demand is staggering. Therefore, the energy demand prediction has become an essential in order to ensure the stable power supply and to meet the future global energy needs. An accurate power consumption (PC) prediction for buildings using the classical methods is a challenging task. In particular, the dependency of PC modelling on various complex factors and the nonlinearity which exists in buildings systems make an accurate prediction of PC difficult. This research presents an approach for accurately predicting power consumption (PC) in buildings using a black-box model based energy demand prediction algorithm. The algorithm utilizes artificial intelligence and a data-driven approach to address the challenges associated with PC modeling, including complex dependencies and nonlinearity in building systems. Two specific models, a Bi-LSTM based model and a hybrid CNN with Bi-LSTM based model are proposed. Further, the impact of various data normalization techniques on model performance is investigated. Results show an improvement of up to 10
^%
in PC prediction when compared to existing LSTM based techniques, highlighting the potential for more robust energy consumption modeling and optimization for energy efficiency.