Estimating envelope thermal characteristics from single point in time thermal images

摘要

Energy efficiency programs implemented nationally in the U.S. by utilities have rendered savings which have cost on average $0.03/kWh. This cost is still well below generation costs. However, as the lowest cost energy efficiency measures are adopted, this the cost effectiveness of further investment declines. Thus there is a need to more effectively find the most opportunities for savings regionally and nationally, so that the greatest cost effectiveness in implementing energy efficiency can be achieved. Integral to this process.;are at scale energy audits. However, on-site building energy audits process are expensive, in the range of US$1.29/m2-$5.37/m2 and there are an insufficient number of professionals to perform the audits. Energy audits that can be conducted at-scale and at low cost are needed. Research is presented that addresses at community-wide scales characterization of building envelope thermal characteristics via drive-by and fly-over GPS linked thermal imaging. A central question drives this research: Can single point-in-time thermal images be used to infer U-values and thermal capacitances of walls and roofs? Previous efforts to use thermal images to estimate U-values have been limited to rare steady exterior weather conditions. The approaches posed here are based upon the development two models first is a dynamic model of a building envelope component with unknown U-value and thermal capacitance. The weather conditions prior to the thermal image are used as inputs to the model. The model is solved to determine the exterior surface temperature, ultimately predicted the temperature at the thermal measurement time. The model U-value and thermal capacitance are tuned in order to force the error between the predicted surface temperature and the measured surface temperature from thermal imaging to be near zero. This model is developed simply to show that such a model cannot be relied upon to accurately estimate the U-value.;The second is a data-based methodology. This approach integrates the exterior surface temperature measurements, historical utility data, and easily accessible or potentially easily accessible housing data. A Random Forest model is developed from a training subset of residences for which the envelope U-value is known. This model is used to predict the envelope U-value for a validation set of houses with unknown U-value. Demonstrated is an ability to estimate the wall/roof U-value with an R-squared value in the range of 0.97 and 0.96 respectively, using as few as 9 and 24 training houses for respectively wall and ceiling U-value estimation.;The implication of this research is significant, offering the possibility of auditing residences remotely at-scale via aerial and drive-by thermal imaging.