Centrality Graph Deep Attention Network and Its Application to Chiller Systems Monitoring

Precise monitoring of chiller systems refers to the real-time detection of equipment or system status using various monitoring techniques. However, traditional state monitoring methods rarely take fault information into account. For this purpose, the centrality graph deep attention network (CenGAT) is proposed for precise monitoring of chiller systems. Firstly, the fault-related variables are selected based on the distribution differences before and after faults to construct the centrality graphs. Then, we utilize graph attention networks to update the centrality graph, which can effectively capture the complex relationships between nodes, thus the nodes with significant influences can be specially monitored. Finally, the autoencoder is carried out to reconstruct the output features to interpret the detected anomalies and enhance the interpretability of the monitoring results.