Novel Physics-Based Machine-Learning Models for Indoor Air Quality Approximations
Published in 29TH ACM SIGKDD Conference On Knowledge Discovery And Data Mining: Mining and Learning from Time Series, 2023
Recommended citation: Mohammadshirazi, A., Nadafian, A., Monsefi, A. K., Rafiei, M. H., & Ramnath, R. (2023). Novel Physics-Based Machine-Learning Models for Indoor Air Quality Approximations. arXiv preprint arXiv:2308.01438. https://arxiv.org/abs/2308.01438
Summary
This paper introduces cost-effective sensors for real-time monitoring of air quality and various indoor conditions, alongside six novel physics-based machine learning (ML) models for precise indoor pollutant concentration predictions. These models integrate state-space concepts in physics with Gated Recurrent Units and Decomposition techniques. Tested in five California commercial building offices, the models prove to be less complex, computationally efficient, and more accurate than comparable state-of-the-art methods. Their success is attributed to a lightweight architecture and the ability to accurately interpret nonlinear patterns in contaminated sensor-collected indoor air quality data.
Authors
Ahmad Mohammadshirazi, Aida Nadafian, Amin Karimi Monsef,Mohammad H. Rafiei, Rajiv Ramnath