Long-Term and Short-Term Energy Prediction using BIM Energy Simulations and Machine Learning Techniques

Abstract

Buildings consume the largest share of electricity in a country’s power grid. There is an ongoing challenge to reduce the energy consumption of buildings. One solution is efficient management of the different systems in the building such as HVAC systems to reduce energy consumption while maintaining the comfort level of people. For proper automation of the systems, there is a need to forecast the energy consumption of a building both in the long term and the short term. The Professor Sir Edouard Lim Fat Engineering Tower building was modelled on Autodesk Revit and energy simulations were performed on the model using Autodesk Green Building Studio. The simulation results were then compared to actual energy consumption. The same building was also used to test machine learning techniques; Gradient Boosting Machine, Support Vector Machine and Deep Neural Network ability to perform short-term energy prediction using data about energy consumption, weather and ambient environment of the building. It was observed that energy simulations overestimated the actual energy consumption by 27%, 29.6%, 59.7% and 60.6% for the months of October, November, December and January respectively. On the machine learning side, Gradient Boosting was observed to outperform SVM and DNN in training time, RMSE and Coefficient of Determination.