Huang, Yuxin, Coursey, Austin, Quinones-Grueiro, Marcos, & Biswas, Gautam. (2024). Time-series few shot anomaly detection for HVAC systems. In Proceedings of the 12th IFAC Symposium on Fault Detection, Supervision and Safety for Technical Processes (SAFEPROCESS 2024), Ferrara, Italy, June 4-7, 2024, Volume 58, Issue 4, Pages 426-431. https://doi.org/10.1016/j.ifacol.2024.07.255
This study addresses a common challenge in detecting anomalies in building heating, ventilation, and air conditioning (HVAC) systems: the limited availability of labeled data needed to train deep learning algorithms. Since labeled data is often scarce, the authors focus on developing a method that can effectively reuse existing data across different systems.
They propose a few-shot domain adaptation approach based on a long short-term memory (LSTM) autoencoder (AE) neural network. This method only requires data from normal system operations (nominal instances) from both a source domain (where more data is available) and a target domain (where data is limited). With just a small amount of data from the target system, the model adapts and improves its performance in detecting anomalies.
The results show that this approach performs better than existing unsupervised methods in various fault scenarios, making it a promising solution for detecting HVAC system anomalies in situations where labeled data is minimal. This method provides a more efficient way to detect system faults while overcoming the data limitations that often hinder the deployment of deep learning models in real-world applications.
