Disaggregation of total energy use into space heating and domestic hot water: A city-scale suited approach

This paper develops a computationally fast algorithm to disaggregate the total energy use recorded by smart heat meters into space heating (SH) and domestic hot water (DHW). The algorithm trains a regression model on SH-only hours and predicts SH for all other hours with potential DHW use. Data from smart water meters were used to identify hours with only SH. Assessing 13 regression models, an untuned random forest was identified as the best-performing regression model with an acceptable computational cost (median: 7.4s per building). Furthermore, using one year of data from over 2400 single-family homes, it was shown that the best result (median CVRMSE: 0.13) can be obtained using only smart heat meters based regressors, increasing the applicability of the developed method. Furthermore, two tests were implemented, and it was successfully demonstrated that they identify situations where the regressors are distributed differently for hours with SH only and hours with SH and DHW, to identify possible training bias; thus, buildings where the disaggregation is potentially unreliable. Validation against data from three single-family houses with known SH and DHW use confirmed the good performance of the method and that the performance can be estimated without ground truth data via nested cross-validation.