A Fault Location and Diagnosis Approach for Distribution Communication Networks

The stability of the distribution communication network is crucial for safeguarding socio-economic activities. With the continuous expansion and increasing complexity of the distribution communication network, traditional fault location and diagnosis methods, which largely rely on engineers' experience and analysis based on fault phenomena and operating conditions. These methods are not only time-consuming and labor-intensive but also fail to achieve comprehensive coverage, which cannot meet the requirements for efficiency and precision. To address these challenges, this paper proposes a fault location and diagnosis algorithm for distribution communication networks. An artificial neural network (ANN) with multiple hidden layers is constructed to learn and extract node information in the switch network, such as traffic, bandwidth, and packet loss. A Graph Attention Network (GAT) is employed as the graph autoencoder to consider the topology relationship of switch nodes. Graph encoder is coupled with the ANN to facilitate attribute learning of the switch network. Then fault risk areas of equipment nodes are divided through the graph clustering model above. For high-risk nodes, a Domain Adversarial Neural Network (DANN) is used to learn fault features from the source domain data and apply them to unlabeled distribution communication network data to achieve fault classification. Simulation results show that the proposed method can effectively learn more distinguishable feature representations of equipment nodes in the distribution communication network. Moreover, through transfer learning, it can efficiently acquire fault feature representations for accurate fault location and diagnosis.