Fault Prediction for Optical Network Based on Hybrid Neural Networks

Optical networks are the cornerstone of the modern information society, serving as the backbone of global information transmission. The consequences of their failure would be significant, resulting in substantial losses. As the complexity and scale of optical networks continue to increase, the maintenance and fault-detection become increasingly challenging. Therefore this paper propose a fault prediction model based on CNN-LSTM-Attention: Firstly, the alarm log is preprocessed, and a sliding time window is employed to convert it into a standard input format for neural networks. Subsequently, the features are extracted through the CNN layer, and their forward and backward temporal relationships are learned through the LSTM layer and the Attention layer. Finally, the output prediction results are generated through the fully connected layer. To validate the efficiency of the proposed approach, a real dataset was employed for evaluation. To demonstrate the superiority of the CNN-LSTM-Attention model, its fault prediction outcomes were compared with CNN and CNN-LSTM. The results demonstrate that the CNN-LSTM-Attention model significantly outperforms the other two models in predicting faults in optical networks.