AN EXPANDED AND UPDATED FRAMEWORK FOR WHOLE-FACILITY ENERGY CONSUMPTION STATISTICAL MODELS OF COMMERCIAL BUILDINGS

Abstract

A significant portion of energy consumption occurs in buildings today. Accurate and easy-to-implement methods are needed to calculate building energy consumption for a wide range of applications, including efficiency assessment, consumption projection, and measurement and verification, to name a few. There are a number of approaches for building energy estimation but the statistical methods have remained popular. As the availability and quality of building energy data continue to improve, the methodologies behind building energy calculation also require updates. This work proposes three new technologies to bring contemporary mindsets to the application of whole-building energy consumption statistical models. The first is a specialised model formulation for the heating hot water consumption for commercial buildings with constant volume reheat systems. It has been observed that the heating consumption of this system type has an unexpected local increase with an increase in ambient temperature caused by dehumidification and reheat. The proposed new method can improve model fit with statistical significance and remove the local trend in the residuals. The second is the use of domestic cold water use or non-HVAC electricity use as an occupancy proxy for building energy models. It is found that combining domestic cold water use with a clustering technique was able to improve model fit by 2.9 percentage points of the CV-RMSE, on average, on top of 14.2% from the traditional weekday-and-weekend method. In the study it was found that other methods, i.e., the use of electricity use as occupancy proxy and the additional of a linear term, were not able to improve the model fit consistently. Finally, a procedure was proposed to examine all data separation or grouping possibilities automatically and comprehensively with pre-defined elementary day-types through a series of lack-of-fit F-tests. This procedure suggests a best separation that balances between model accuracy and simplicity. It was tested on measured energy consumption data of 76 case study commercial buildings. The proposed method weighed simplicity more heavily than traditional statistical complexity-penalising metrics. The new method improved the CV-RMSE by 7.6 percentage points, on average, and helped extract information to help better understand the buildings’ energy consumption patterns.