From greenhouse gas emission reductions to the implementation of energy efficiency standards and the electrification of homes, government authorities must adopt ambitious measures to counter the looming climate disruptions. Understanding the long-term impacts of these actions is therefore critical.

A number of energy models, including the Canadian TIMES model generator, are able to predict national energy consumption. While they help governments shape their economic and environmental policies, they provide a comprehensive view that does not allow for accurate smaller-scale projections. Yet, for a city, for example, reducing its energy consumption remains a particular challenge. Is there a way to predict the effects at a local level?

This very question is the focus of the research conducted by Adam Neale, PhD student in mechanical engineering at Polytechnique Montréal. With the support of supervisors Michel Bernier and Michaël Kummert, with whom we collaborate, he is developing a bottom-up energy model based on individual buildings to make small-scale predictions. Together, the researchers came up with the idea of relying on tools already installed in a vast majority of homes: communicating meters that transmit a range of data every 15 minutes. But these data are confidential. Should they be granted access once all the ethical issues have been addressed, the experts will be inundated with totally anonymous big data.

To establish and test the model, the team had to simulate the information. Engineers modeled 200 000 virtual homes and generated data that meters would likely transmit. Each home was attributed different characteristics based on information available in the literature. By varying parameters including the area to heat or resistance of walls and insulation, it was possible to define the link between these variable and the electrical power.

To perform the complex calculations, the scientists relied on a powerful server that took just 48 hours to simulate data from the 200 000 virtual homes. Each simulation served to train and refine the model, and machine learning made it possible to determine the type of heating system in a home, its level of insulation and even the number of household members!

Most current energy models only provide a big picture. They aren’t designed to model a subset of buildings, since they take a top-down approach. Our vision is bottom-up: working toward something based on individual buildings.”

– Adam Neale, Ph.D. Student at Polytechnique Montréal

Thanks to these rigorous data, organizations will now be able to assess the impact of the construction of a new neighborhood on the distribution systems or predict the energy savings following the implementation of more rigorous insulation standards to the nearest hour.

The model could eventually apply in any part of the world. Professor Bernier aims to make the model and dataset from the virtual meters available in open source to all members of the scientific and industrial communities.