| Abstract | This report is the final deliverable of the project A1-017320.1. The main purpose of this report is to investigate the feasibility of using one of the emerging data sources, i.e. smart thermostats, along with machine learning approaches to detect and quantify overheating in buildings. This report is organized in two sections. Section A represents the results of the indoor overheating analysis in residential buildings based on smart thermostat data. In section B, a data-driven indoor conditions warning framework for residential buildings is developed and assessed.
In section A, indoor temperature and humidity measurements from more than 3,000 connected thermostats, during summers of 2016-2019 and from homes located in five major metropolitan areas in Canada were used. Furthermore, a robust procedure to evaluate the frequency and severity of indoor overheating based on the heat-related health outcomes for older people was utilized. In particular, the overheating occurrence in homes with and without central air conditioning (AC) units was comparatively evaluated. The results of the first section showed that 12% of houses under study experienced at least one overheating event. Furthermore, the extreme indoor overheating events with potential health risks to the vulnerable occupants were more common amongst houses without central air conditioning units across the selected cities.
In section B, a feasibility study was done on more than 180 houses located in Ontario that do not use central AC units. In this approach, two gray box models were developed based on the hourly indoor and outdoor temperatures. The gray box models were implemented in a recursive forecasting strategy using sliding training and forecast time periods to provide the forecasted hourly indoor temperatures. The proposed models were able to forecast 12-hour ahead hourly indoor conditions in 92% of houses with less than 5% mean absolute percentage error. This indicated the potential for leveraging data from IoT devices and machine learning, as part of an overheating detection and warning system in buildings. Although this work was based on data from residential buildings, the methodologies developed here can be used for other buildings. |
|---|
