Estimating Air Exchange Rate using Concentration Time Series Data based on Unsupervised Learning

Bowen Du, JEFFREY SIEGEL, University of Toronto

     Abstract Number: 3
     Working Group: Indoor Aerosols

Abstract
Air exchange rate affects indoor air quality and building energy consumption. This building parameter is highly variable due to the variations of weather and building operation, while conventional tracer gas methods only provide a snapshot of the building ventilation condition as they are intrusive and time-consuming, leaving a critical knowledge gap in the long-term air exchange rate of buildings of different types. In this study, we used an unsupervised machine learning model to automatically recognize decay trends from pollutant concentration time series and estimate building air exchange rate based on mass balance. The model uses k-means clustering for extracting decay periods and then density-based spatial clustering of applications with noise (DBSCAN) for grouping them. We tested the feasibility and generalizability of the model on multiple datasets collected from various university and residential environments. Results suggested that the proposed model can recognize several decay episodes per day based on CO₂ and PM_2.5 concentration data. The estimated decay rate of CO₂ is consistently lower than that of PM_2.5 as the former is a proxy of air exchange rate while the latter is a proxy of effective air exchange rate including deposition and potential filtration. Further, the decay rate of both varies spatially and temporally. The impact of clustering hyperparameters, baseline concentrations, sensor accuracy, and result filtration on the estimated air exchange rate is discussed in detail to develop a protocol for selecting the optimal model parameters and ensuring reproducibility. Overall, the proposed model provides a low-cost solution to monitoring long-term air exchange rate, which also has potentially wide applications including assessing air mixing conditions and characterizing the emitting profile of indoor air pollution sources.