Modeling of Fault Propagation Paths in Power Communication Networks Combined with GNN

Power communication network faults often propagate with changes in node status and network topology, and their significant dynamic evolution characteristics are difficult to capture effectively. Capturing these dynamically evolving faults is critical to guaranteeing the reliability of power communication links and continuous industrial operation. To this end, this paper proposes a method combining graph neural networks (GNNs) to construct a propagation path modeling framework that integrates topological structure and node status time series characteristics. A time-slice network snapshot is constructed based on the node status and communication link. The temporal feature extraction is enhanced by position encoding and sliding time window. The T-GCN (Temporal Graph Convolutional Network) model is combined with the graph topology and dynamic propagation mode to predict the failure probability of each node in the power communication network and generate a propagation path diagram. Finally, the attention mechanism is used to identify key nodes, and the actual propagation trajectory is used to supervise the training to achieve accurate early warning of potential paths. Experiments show that the proposed model has a precision of 0.92 in node fault prediction, an average time error of only 1.2 minutes, a key node coverage of 90.5%, a propagation delay reduction of 31.2%, an average coverage of 93.2%, and a minimum F1-score of 0.84 in diverse fault scenarios, demonstrating high precision and strong generalization ability. Because the method models node state, topology and propagation delay, it is directly applicable to reliability analysis of electromagnetic communication links in power networks.