Cross comparison of empirical and simulated models for calculating residential electricity consumption

• We develop air conditioning demand models from empirical and estimated data.
• We compare the statistical significance across models and across datasets.
• Accounting for seasonality yielded stronger models.
• Engineering values, available in home audits, provide more accurate models.
• Other indicators (directionality and shading) surface as statistically significant.

The U.S. Energy Information Administration's Residential Energy Consumption Survey (RECS) provides electricity end-use estimates derived from disaggregating total household consumption. However, the accuracy of the EIA's disaggregation method is unclear, thus potentially producing erroneous end-use estimates and affecting respective decision support. In order to test the EIA's disaggregation methods, we perform parallel empirical analysis of residential cooling demands on two household samples located in Texas: an estimated sample (RECS) and a sample where air conditioning branch circuits are directly metered. Results indicate statistically significant differences between data sources, with the RECS sample possibly underestimating cooling demands. Results further indicate that models developed using observed end-use consumption are more robust, demonstrating statistically significant predictors that do not significantly correlate with end-use estimates from the RECS sample. We find that accounting for seasonality increases predicted R2 values and reduces overestimation of statistical significance. Finally, the observed model indicates that window area and orientation are statistically significant predictors of cooling loads, which are parameters not surveyed in RECS. We then provide policy recommendations to improve end-use estimates, including integrating select utility-funded audit data into disaggregation methods.