Research On Alarm Correlation Analysis And Fault Prediction Of Communication Networks

With the rapid development of China's communication technology,the scale of communication networks continues to expand,and the complexity also increases.Any equipment fault in the network may cause associated equipment to generate alarms,which will trigger more and more alarms,resulting in the personnel of communication network management system unable to locate faults in a timely and accurate manner,which brings great difficulties to the maintenance of communication networks.The fault prediction of communication networks can timely troubleshoot network problems,improve the efficiency of network operation and maintenance,and is of vital significance to the management of communication networks.In order to solve the problems of poor flexibility and low precision in traditional communication network fault prediction methods,this thesis applies alarm correlation analysis to network fault prediction and carries out research work on alarm correlation analysis and fault prediction of communication networks.The topic is selected from the "Key Customer Alarm Data Analysis" project of China Telecom System Integration Co.,Ltd.The main contents of this thesis include:1)Aiming at the problem that traditional communication network alarm correlation analysis methods cannot adapt to the complex and changeable network structure,an alarm correlation analysis method based on data mining is proposed.This thesis assigns flexible weights to alarms based on the alarm freshness,that is,the alarm occurrence time,and proposes Dynamic Weighted Sequential Pattern Mining(DWSPM)algorithm to analyses the correlation of communication network alarms.The weight of alarms can be adjusted adaptively according to the alarm freshness.In addition,a minimum weight threshold is introduced to constrain the pruning process of the DWSPM algorithm.The experimental results show that the DWSPM algorithm can greatly reduce the execution time and improve the execution efficiency,providing a reliable dataset for the communication network fault prediction model.2)In order to solve the problem that the existing communication network fault prediction methods are lack of deep data mining,a communication network fault prediction method based on sample equalization and feature interaction is proposed.This thesis uses Wasserstein Generative Adversarial Networks-Gradient Penalty(WGAN-GP)to generate new minority samples to solves the problem of sample imbalance in the alarm dataset.Moreover,this thesis proposes a Memory based Feature Generation by Convolutional Neural Network(M-FGCNN)model for communication network fault prediction.The model is based on Multi-Layer perceptron(MLP)and Convolutional Neural Networks(CNN)to strengthen the interaction between features.The experience of experts in the warning field is used to generate new warning features based on the Factorization Machine(FM)model.At the same time,the memory vector is added into the embedding matrix and the loss function is modified to enhance the memory of the model.Through experiments on the public unbalanced dataset,it is verified that the WGAN-GP model can generate new data with better quality than the existing methods.Furthermore,it is verified that the proposed M-FGCNN model has better fault prediction performance compared with other deep learning models on real alarm datasets.