Simulating a Nationally Representative Housing Sample Using EnergyPlus
This report presents a new simulation tool under development at Lawrence Berkeley National Laboratory (LBNL). This tool uses EnergyPlus to simulate each single-family home in the Residential Energy Consumption Survey (RECS), and generates a calibrated, nationally representative set of simulated homes whose energy use is statistically indistinguishable from the energy use of the single-family homes in the RECS sample. This research builds upon earlier work by Ritchard et al. for the Gas Research Institute and Huang et al. for LBNL. A representative national sample allows us to evaluate the variance in energy use between individual homes, regions, or other subsamples; using this tool, we can also evaluate how that variance affects the impacts of potential policies.
The RECS contains information regarding the construction and location of each sampled home, as well as its appliances and other energy-using equipment. We combined this data with the home simulation prototypes developed by Huang et al. to simulate homes that match the RECS sample wherever possible. Where data was not available, we used distributions, calibrated using the RECS energy use data. Each home was assigned a best-fit location for the purposes of weather and some construction characteristics.
RECS provides some detail on the type and age of heating, ventilation, and air-conditioning (HVAC) equipment in each home; we developed EnergyPlus models capable of reproducing the variety of technologies and efficiencies represented in the national sample. This includes electric, gas, and oil furnaces, central and window air conditioners, central heat pumps, and baseboard heaters. We also developed a model of duct system performance, based on in-home measurements, and integrated this with fan performance to capture the energy use of single- and variable-speed furnace fans, as well as the interaction of duct and fan performance with the efficiency of heating and cooling equipment. Comparison with RECS revealed that EnergyPlus did not capture the heating-side behavior of heat pumps particularly accurately, and that our simple oil furnace and boiler models needed significant recalibration to fit with RECS.
Simulating the full RECS sample on a single computer would take many hours, so we used the "cloud computing" services provided by Amazon.com to simulate dozens of homes at once. This enabled us to simulate the full RECS sample, including multiple versions of each home to evaluate the impact of marginal changes, in less than 3 hours.
Once the tool was calibrated, we were able to address several policy questions. We made a simple measurement of the heat replacement effect and showed that the net effect of heat replacement on primary energy use is likely to be less than 5%, relative to appliance-only measures of energy savings. Fuel switching could be significant, however. We also evaluated the national and regional impacts of a variety of "overnight" changes in building characteristics or occupant behavior, including lighting, home insulation and sealing, HVAC system efficiency, and thermostat settings. For example, our model shows that the combination of increased home insulation and better sealed building shells could reduce residential natural gas use by 34.5% and electricity use by 6.5%, and a 1 degree rise in summer thermostat settings could save 2.1% of home electricity use. These results vary by region, and we present results for each U.S. Census division.
We conclude by offering proposals for future work to improve the tool. Some proposed future work includes: comparing the simulated energy use data with the monthly RECS bill data; better capturing the variation in behavior between households, especially as it relates to occupancy and schedules; improving the characterization of recent construction and its regional variation; and extending the general framework of this simulation tool to capture multifamily housing units, such as apartment buildings.