Inverse energy model development via high-dimensional data analysis and sub-metering priority in building data monitoring

Abstract In the US, building sector consumes approximately 41% of all U.S. primary energy and 70% of all generated electricity. Office building uses the largest percentage of primary and derived energy in the commercial buildings sector. Therefore, energy saving is the primary target of building renovation and equipment retrofit. To estimate energy savings after a renovation or equipment retrofit, measurement and verification (M&V) must be conducted based on energy use models or benchmark tools to track and assess savings. Although some methods and modeling tools are applied more often, there is no well-accepted model formulation methodology and some methods have high uncertainty. This paper discusses a method to formulate the energy use model as an inverse regression model via an advanced and robust high-dimensional data analysis method. According to the inverse regression model, independent variables determined to be most influential to the energy consumption should be metered prior to building retrofit or audit. To select the most influential variables, this research considers a comprehensive set of potential key variables to avoid underspecified model issue. The models established by this method are observed to have great power in predicting the building energy consumption. More importantly, through the variable selection procedure, the method identifies key variables that should be monitored to continuously improve the building energy performance. A simulation study and a case study are presented to show the effectiveness of the new approach.