Machine learning, neural networks

Detecting coolant leaks using a graph neural network

S. S. Ivanov, M. V. Ronkin

Ural Federal University named after the First President of Russia B. N. Yeltsin, Ekaterinburg

Abstract: The aim of this study is to develop a method for detecting refrigerant leaks in a data center using a modern graph artificial neural network capable of processing inter-feature and temporal dependencies. The graph neural network is compared with other popular methods for time series prediction and anomaly detection that were tested on the same dataset. Experimental results demonstrate the greater effectiveness of the graph neural network. The model's robustness is tested using experiments incorporating white noise into the time series signals. The dataset is collected at an industrial container-type data center.

Keywords: data center, refrigerant leak, neural networks, graph neural networks, time series, anomaly detection.

DOI: 10.14357/20718594250409

Bibliographic databases:

•