Virtual home energy auditing at scale: Predicting residential energy efficiency using publicly available data

In this study we model and examine the energy efficiency profile of individual single-family houses from Gainesville, Florida, in our sample (n = 7091). For this we use Princeton Scorekeeping Method (PRISM) which processes historical weather data and monthly utility usage data as inputs using an iterative regression approach to compute three energy efficiency parameters: (1) baseload consumption for end-uses which do not change with weather, e.g., lighting, refrigerator, water heater; (2) heating/cooling slope which is a function of the building shell insulation and the efficiency of the heating/cooling unit; (3) reference temperature, i.e., the outside temperature at which the house turns on heating/cooling. These parameters make up the normalized annual consumption (NAC). We then proceed to regress these parameters against the publicly available data to study the extent we can extract statistical insight for residential energy efficiency profiling using publicly available information (n = 5243). These regression models are to pave a path to creating energy efficiency “reservoir maps” across individual homes and reducing the information barrier to energy efficiency adoption.