A. Dahe1, V.V. Stuchilin2
1, 2 National University of Science and Technology «MISIS» (Moscow, Russia)

1 m2112269@edu.misis.ru, 2 stuchilin.vv@misis.ru

The problem of timely detection of malfunctions in such systems is very important, since failures can negatively affect the comfort and safety of residents, as well as lead to significant repair and maintenance costs. The aim of the study is to create a synthetic dataset for machine learning models, which will be used as a reference for detecting and predicting defects in the life support systems of residential buildings. As part of the study, a synthetic dataset has been created that includes key parameters of the life support system, such as temperature, pressure, humidity, CO2 levels and energy consumption, taking into account possible changes in operating conditions. In addition, noise has been introduced into the dataset to simulate real measurement errors, which increased the reliability of testing machine learning models. Fault scenarios have been generated based on a probabilistic approach that models the relationship between various system parameters. After receiving the data, detailed statistical analysis and preprocessing have been carried out, including outlier processing, class balancing, and parameter normalization to ensure their suitability for model training. The results showed that the proposed method of data generation and analysis makes it possible to effectively simulate the operating and emergency conditions of the system, which makes it suitable for use in the development and testing of fault detection and prediction models. The practical significance of the work lies in the creation of a set of data that can be used to develop intelligent monitoring and diagnostic systems in real-world operating conditions of building life support systems, and ultimately ensure the continuous and efficient operation of building life support systems.

Dahe A., Stuchilin V.V. Data modeling for machine learning-based fault detection and prediction in building life support systems. Neurocomputers. 2025. V. 27. № 3. P. 73–81. DOI: https://doi.org/10.18127/j19998554-202503-09 (in Russian)

1. Schiendorfer A., Zimmermann G., Lu Y. et al. Fault diagnosis in HVAC systems based on the heat flow model. 2014. DOI: 10.13140/ 2.1.4189.0249.
2. Yan Y., Luh P.B., Pattipati K.R. Fault prognosis of key components in HVAC air-handling systems at component and system levels. IEEE Transactions on Automation Science and Engineering. 2020. V. 17. № 4. P. 2145–2153. DOI: 10.1109/TASE.2020.2979166.
3. Tun W., Wong K.W., Ling S.H. Advancing fault detection in HVAC systems: Unifying Gramian angular field and 2D deep convolutional neural networks for enhanced performance. Sensors. 2023. V. 23. № 18. DOI: 10.3390/s23187690.
4. Patent № 10592821 US. Self-learning fault detection for HVAC systems. D.E. Denton, C.L. Garrett. March 17, 2020.
5. Zhang B., Wang P., Liu G. et al. AHU sensor fault diagnosis in various operating conditions based on a hybrid data-driven model combined energy consumption. Journal of Building Engineering. 2024. V. 87. P. 109028. DOI: 10.1016/J.JOBE.2024.109028.
6. Kiamanesh B., Behravan A., Obermaisser R. Fault injection with multiple fault patterns for experimental evaluation of demand-controlled ventilation and heating systems. Sensors. 2022. V. 22. № 21. DOI: 10.3390/s22218180.
7. Fan S., Zhang X., Song Z. Imbalanced sample selection with deep reinforcement learning for fault diagnosis. IEEE Transactions on Industrial Informatics. 2022. V. 18. № 4. P. 2518–2527. DOI: 10.1109/TII.2021.3100284.
8. Liu J. Importance-SMOTE: a synthetic minority oversampling method for noisy imbalanced data. Soft Computing. 2021. V. 26. № 3. P. 1–23. DOI: 10.1007/S00500-021-06532-4.
9. Deng M., Guo Y., Wang C., Wu F. An oversampling method for multi-class imbalanced data based on composite weights. PLoS One. 2021. V. 16. № 11. DOI: 10.1371/JOURNAL.PONE.0259227.
10. Ding H., Xu J. Random gradient descent tree: A combinatorial approach for SVM with outliers. Proceedings of the AAAI Conference on Artificial Intelligence. 2015. V. 29. № 1. P. 2561–2567. DOI: 10.1609/AAAI.V29I1.9571.
11. Liang J., Du R. Model-based fault detection and diagnosis of HVAC systems using support vector machine method. International Journal of Refrigeration. 2007. V. 30. № 6. P. 1104–1114. DOI: 10.1016/j.ijrefrig.2006.12.012.
12. Liu J. Importance-SMOTE: a synthetic minority oversampling method for noisy imbalanced data. Soft Computing. 2022. V. 26. № 3. P. 1141–1163. DOI: 10.1007/S00500-021-06532-4/TABLES/9.
13. Weinberger K.Q., Saul L.K. Distance metric learning for large margin nearest neighbor classification. Journal of Machine Learning Research. 2009. V. 10. P. 207–244.
14. Shaw B., Huang B., Jebara T. Learning a distance metric from a network. Advances in Neural Information Processing Systems. 2011. V. 24. P. 1899–1907.