Residential Thermal Performance Analytics

A desktop study was performed to evaluate the accuracy and usefulness of an online service that infers home thermal performance characteristics from electricity consumption (from Green Button data) and weather (from the vendor’s proprietary network of stations). The service provides customers with a monthly report rating the weather-readiness of their home and the efficiency of their energy use profiles. No onsite audit is performed, and minimal details (such as home location and floor area) are collected upon enrollment. This study evaluated the accuracy of home performance scores in four categories: Insulation, Solar Resistance1, Draftiness, and Overall. Scores were compared to actual home characteristics sourced from the Energy Upgrade California (EUC) database of several hundreds of energy audits.

For the dataset of 97 homes in San Diego Gas & Electric Company (SDG&E) territory, no consistent correlations were found between the performance scores and actual home characteristics from the EUC database, “EUC metrics.” The linear regressions had high uncertainty, due to small sample size. (The 97 homes were divided into nine groups, per the vendor’s methodology for calculating scores relative to similar homes.) The EUC metrics were another source of uncertainty. (Only one was directly comparable to the corresponding performance score.) The vendor was able to correctly identify two out of ten energy upgrades, but had high false positive and false negative rates. Some evidence suggested the vendor’s disaggregation method overestimates heating, ventilation, and air conditioning (HVAC) energy use in hot months, but the evidence was based on questionable EUC data. Further verification is advised.

The monthly report was found to be generally useful, with some suggested improvements. Specifically, the Overall score was found to be a good indicator of total electricity use. Suggested improvements include emphasizing general statements of weather-related energy use and associated recommendations instead of providing numerical scores for home characteristics. It is also suggested to increase sample size by not binning comparison groups by floor area.