Fault Diagnosis and Location based on Graph Neural Network in Telecom Networks

As the scale of telecom networks continues to expand and the structure becomes more complex, telecom networks generate a large number of continuous warning messages every day. It may lead to massive network paralysis. To solve this problem, this paper proposes a fault diagnosis scheme of telecom networks based on Graph Neural Network (GNN). The scheme can be applied to locate the fault-root-devices. Firstly, Long Short-Term Memory (LSTM) is used to cluster the state level of partial devices, and the results provide prior information for the construction of the GNN. Then, the embedded representation of graph network is introduced to aggregate the information of the whole graph. Finally, the device network is set up and the data information is embedded into the whole graph network. GNN model is built to diagnose the fault of all devices and locate the fault- root-devices. Moreover, the performance of the model constructed respectively by Graph Convolutional Network (GCN) and Graph Attention Network (GAT) are compared and analyzed. Based on public fault data, the experimental results show that the accuracy of GAT is higher than that of GCN; the correct rate of fault diagnosis can exceed 93 %; and the fault- root-devices can be accurately located. In addition, the complexity is only O(n) in our scheme.