Application of supervised learning methods to better predict building energy performance

Building energy consumption is shaped by a variety of factors which prompts a challenge of accurately predicting the building energy performance. Research findings disclosed a significant gap between the building’s predicted and actual energy performance. One of the key factors behind this gap is the occupant’s behavior during operation which includes a set of dependent and independent parameters generating a greater level of uncertainties. To accurately estimate the energy performance, we need to quantify the impact of any observed parameters and further detect its correlation with other parameters. Human behaviors are complex and quantifying the impact of all its interconnected parameters can be error prone and costly.
To minimize the performance gap, more scalable and accurate prediction approaches, such as supervised machine learning methods, should be considered.
This paper is devoted to investigate the most commonly used supervised learning methods which, when intertwined with conventional building energy performance prediction model, could potentially provide more accurate and reliable estimates. The paper will pinpoint the best use of each studied method in the relation to energy prediction in general and occupant’s behavior in specific and how it can be implemented to better predict building energy performance.