HVAC system runtime is a critical factor in building energy use and indoor air quality (IAQ). While there are well-documented methods for measuring HVAC runtime in residential buildings by monitoring supply air temperature and its fluctuations over time (e.g., Li et al., 2018; Thornberg et al., 2004), there are currently no known studies that directly apply machine learning (ML) to time-series indoor environmental quality (IEQ) data (e.g., temperature and relative humidity) to predict runtime. Using IEQ data in an ML model to estimate HVAC runtime offers several advantages: model diversity, scalability, simplicity, and feature importance insights that are not readily available through conventional runtime measurements. In particular, with the rise of Big Data, scalability is crucial for processing increasingly large datasets effectively, and a pre-trained ML model can help handling IEQ data in a facilitated way.
This repository aims to provide a working prototype for ML development to estimate HVAC runtime in a single-family home located in Toronto, Ontario. This work is a side project of the Mahdavi et al. (2020) EP, where our team introduced quantitative filter forensics (QFF), a novel approach for indoor air sampling that depends heavily on accurate HVAC runtime data. A high-accuracy runtime estimation algorithm would be a valuable asset for advancing QFF and indoor air sampling methods. A link to the full article is provided in the "About" section.
In this project, labeled runtime data was gathered using a pressure sensor in the HVAC supply duct of the studied home, serving as the target variable for ML models. Temperature, relative humidity, particle concentration, and other IEQ parameters were collected over approximately six weeks during summer. With a complete set of features and target variables, various ML algorithms—such as logistic regression, support vector machine (SVM), random forest, and XGBoost—were employed to explore how different models impact overall accuracy in predicting HVAC runtime.