OTN Fault Localization Algorithm Based on Graph Neural Networks

With the rapid development of the telecommunications industry, the network rate interface of Optical Transport Network (OTN) can reach 400G, becoming the core backbone for carrying massive data and service signals. A single network fault, such as node or link failure, can severely threaten Quality of Service (QoS) and lead to service interruptions. Therefore, rapid and accurate fault localization and repair are crucial. To overcome the current issues faced by OTN fault localization technologies, such as low accuracy and insufficient dynamic adaptability, we propose an OTN fault localization algorithm based on graph neural networks (GNN). First, we construct an alarm knowledge graph (KG) that tightly associates complex alarm information with network device metrics. Subsequently, we utilize GNN to deeply mine potential fault patterns in the network topology, achieving intelligent localization of complex fault scenarios; to capture the causal relationships among alarms, we introduce a causal attention mechanism. Experimental results show that our proposed localization method can quickly identify root alarms during alarm storms, significantly improving the accuracy of fault localization, reducing localization time, and enhancing operational efficiency and service quality in communication networks. This research provides strong technical support for the stable operation of OTN networks and holds significant practical application value and theoretical significance.