Research On Fault Prediction Technology Based On Hybrid Neural Network Model

Today's social science and technology is developing very fast.In people's production and life,network systems have become ubiquitous.People can use the network to improve work efficiency and enrich daily life.Made important contributions.It is precisely because of the increasing role of people on the network,the scale of the network itself is becoming larger,the structure is more and more complex,and the amount of data is also very huge.Therefore,when the network fails,the difficulty of maintenance will increase.Large,if the network takes a long time from failure to normal operation,it will cause great losses to human beings,so how to reduce the loss caused by network failures has become an important topic for people to study.In recent years,more and more researches have been done on fault prediction technology.The fault prediction technology can detect possible network failures in advance,enable people to prepare for network failure repairs in advance,and reduce the time for network recovery work,thereby reducing loss.But most of them are structured data such as alarm data,and unstructured data such as logs are not much studied.Logs are data that the system is generating every moment,so a large number of them are hidden to reflect the system operation.The state information has good research value,so this paper proposes a log-oriented fault prediction technology based on neural network.The main contents are as follows:1.Data preprocessing,because the logs used in this article are unstructured and not as "clean" as structured data,unstructured logs will have some useless symbols and information.Before entering the logs into the predictive model The network log needs to be pre-processed,useless data is washed out,samples are extracted using a sliding window,and the log is converted into a matrix vector form that the fault prediction model can recognize through the word embedding layer.2.CNN-based fault prediction.CNN includes a convolutional layer,a pooling layer,and a fully connected layer.CNN has a good feature extraction capability.In this paper,CNN is used to extract log data features and predict whether the network's future operating status will fail.3.Fault prediction based on CNN-LSTM hybrid neural network.In order to verify whether the performance of the hybrid neural network prediction model is higher than that of a single CNN,this paper introduces LSTM based on the CNN failure prediction model.LSTM has a good prediction ability for sequences.This paper uses CNN-LSTM hybrid neural network Predict the running status of the network and compare the performance of two fault prediction models based on CNN andCNN-LSTM.