A novel fault node diagnosis of interconnected network based on PMC-syndrome-enhanced graph neural network

System-level fault diagnostic models play important roles in improving the reliability of interconnected networks. However, relying solely on traditional system-level diagnostic model is often constrained by the requirement of graph structure regularity and the number limitation of diagnosable nodes. This paper proposes a novel fault diagnosis (TGSEGNN) algorithm based on PMC-syndrome-enhanced graph neural network (SEGNN) and test graph (TG) under the PMC model, which overcomes the above shortcomings with wide applicability and robustness. First, this paper proposes a constructing method of the TG based on the PMC model, which significantly enriches the input feature expression through the feature matrix that combines the test results of neighboring nodes and the syndrome information for target node. Then, an improved SEGNN is designed to embed the PMC-syndrome into the feature aggregation process of neighboring nodes, and the complex relationship between nodes was modeled using multi-layer networks, which enhanced the model’s reliability. Finally, a fault node determination method is proposed, using a fully connected layer and a softmax classifier with a threshold for accurate diagnosis. The experimental results show that TGSEGNN is significantly better than the OGGCN, OGGAT, OGSAGE, TGGCN, TGGAT, and TGSAGE methods in terms of accuracy, precision, recall, F1 score, and AUC.