Predicting how much energy individual homes consume is a persistent challenge for researchers and urban planners. Models often group houses into broad categories based on their age or other characteristics, overlooking differences in their size, construction and energy performance.
A new approach developed by Concordia researchers with colleagues at McGill University, Natural Resources Canada and the National Research Council Canada, could make these predictions considerably more accurate. As described in the journal Energy & Buildings , the model offers a better way to assess energy use across residential neighbourhoods and identify opportunities for reducing consumption.
Their method, called Three-Layer Filtering, progressively refines how buildings are represented in urban building energy modeling. These computer models help researchers understand energy consumption across neighbourhoods and cities, and can inform decisions about building efficiency and decarbonization, the researchers say.
The team tested their model on 100 randomly selected detached houses in Montreal that use electric baseboard heating. They found their approach reduced the average error in predicting individual homes' annual electricity consumption from 80.7 per cent when using existing methods to just 4.3 per cent, providing more accurate data for urban planners and utility officials.
The first step narrows the analysis to comparable buildings, avoiding comparisons between fundamentally different types of homes. The second groups houses according to several characteristics, including construction year, floor area, number of storeys, volume and estimated energy performance, rather than relying on age alone. The houses are then grouped into seven distinct clusters based on their similarities, with each cluster representing an archetype.
In the final step, the researchers adjust the model for individual houses based on their corresponding archetypes. Machine-learning tools help identify the building characteristics that most influence energy consumption, while a simplified version of the model learns from detailed energy simulations to quickly estimate energy consumption under different parameters. Together these measures significantly reduce the computing power needed to make these adjustments.
After the initial grouping, the average prediction error fell to 23.3 per cent. Following individual calibration, 83 per cent of the houses had predictions within five per cent of their measured annual electricity consumption.
This approach avoids having to build and run a separate, resource-intensive model for each property. This could make detailed, building-level energy estimates more practical for municipalities working with large housing stocks, rather than limiting them to small samples or broad neighbourhood averages.
The simplified computational approach offers an additional speed advantage. In a test involving one representative house, the researchers' streamlined model performed almost 22 times faster than conventional simulations, while producing comparable results.
The findings suggest that more accurate, building-specific energy estimates could eventually help municipalities identify homes with high energy consumption, prioritize retrofits and plan the transition to more efficient buildings.
“A city like Montreal is made up of many different neighbourhoods, and each one has a unique energy consumption pattern,” says lead author Wei Zhou , a PhD candidate at the Centre for Zero Energy Building Studies (CZEBS) in the Department of Building, Civil and Environmental Engineering .
“With our model, urban planners can do a speedy and accurate analysis of specific areas and determine which retrofit scenario is most appropriate.”
While the study remains a proof of concept, Zhou says it is scalable to other cities, building stocks, heating systems and climates.
Additional research by Navid Shirzadi of Natural Resources Canada; Dongxue Zhan, Peng Liu, Liangzhu Wang and Andreas Athienitis at CZEBS; Travis Moore, CZEBS and National Research Council Canada; and Shaoxiang Qin, CZEBS and McGill University.
The research received funding from the Mitacs Accelerate, the Natural Sciences and Engineering Research Council of Canada and the Canada First Research Excellence Fund.
Read the paper: “ Progressive filtering and cluster-localized calibration for customized UBEM ”
Energy and Buildings
Computational simulation/modeling
Not applicable
Progressive filtering and cluster-localized calibration for customized UBEM☆
11-Sep-2026
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.