Fault Data Analysis Based On Autoencoder In Optical Network

In recent years,with the continuous development and improvement of communication infrastructure,a variety of new services and applications based on communication technology emerge endlessly,bringing great convenience to people’s work and life.Optical network,as the foundation of communication infrastructure,its health and stability are very important.However,with the deepening of the degree of heterogeneity of the network and the expansion of the scale of the network,its operation and maintenance costs and difficulties are increasing,which brings great challenges to engineers.So network fault prediction is particularly important.If you can accurately predict the faults that will occur on the network,engineers can take actions in advance to avoid network faults.However,traditional optical network fault prediction methods lack of comprehensive analysis of historical data and can not accurately predict network faults.Deep learning is a technology that has caught fire in recent years.It can learn the mapping between input and output from huge amounts of data and perform well in various fields.The application of deep learning to network fault prediction can achieve good results.In this paper,based on the real data from an operator’s network,combined with the deep learning method,the problems of optical network fault prediction and fault data analysis are studied.The main work and innovations are as follows:First,a semi-supervised fault prediction method is proposed to solve the problem of unbalanced data.This method is based on autoencoder and improved to form a new semi-supervised model.The model is based on reconstruction error to predict,and only needs normal data during training,which reduces the dependence on fault data and avoids the problem of data imbalance.In the data set of this paper,the prediction effect is as follows:the false negative rate is 0.0040,the false positive rate is 0.0386,the accuracy rate is 0.9680,and the F1 value is 0.9224.Secondly,PCA algorithm and hidden space data distribution are used to analyze the fault data,and the soft faults are further refined into explicit faults and implicit faults.explicit fault refers to a certain feature or several features obviously beyond the normal range;Implicit fault refers to a device failure when all features are within the normal range.Furthermore,this paper analyzes the causes of the two kinds of faults,which provides a reference for the future fault cause analysis.