Alarm Analysis And Failure Location Technology For Optical Transport Network Based On Machine Learning

As the scale of optical transport network continues to expand,the impact of optical network failures is also increasing.In order to ensure service quality of the optical transport network and reduce the losses caused by network failures,it is necessary to study the failure location technology of the optical transport network.Transport network failures can be divided into hard failures and soft failures.Soft failures caused by the deterioration of the performance parameters have unfavorable impact on network operation and complicate the location technology.Therefore,it is of great significance to study the soft failures in the optical network.The main research contents and innovations of this thesis are as follows:(1)Combine common optical transport network alarm information,analyze the correlation between the cause of the alarm and the failure,and determine the studied type of soft failure.It is proposed to introduce the fiber nonlinear effect into the soft failure type,use the simulation software Opti System to build an optical communication simulation system,and collect two kinds of failure data sets.One of two sets is subject to the fiber nonlinear effect and the other is not.These two sets are used to train the failure location model.(2)It is proposed to use the coordinate point information of the constellation diagram to monitor the type of soft failures,and extract the features of the constellation diagram information by adopting the improved K-Means algorithm.This method not only avoids the drawbacks of manually extracting features,reduces the time cost,improves the accuracy of failure location,but also reduces the complexity of the algorithm for failure location using constellation images,and saves memory space.A failure location algorithm based on the improved K-Means clustering combination is proposed,which combines Decision Tree,SVM and 1DCNN algorithms.The results which are obtained from three algorithm models of K-DT,K-SVM and K-1DCNN are compared and analyzed.The performance of the K-SVM failure location model is the best.(3)On the basis of a single algorithm model,the idea of integrated learning is introduced.Firstly,simulation experiments on RF and the XGBoost failure location algorithm model based on Bayesian optimization are performed.Compare this model with a single Decision Tree model and analyze the compared results.Subsequently,the failure location algorithm based on multi-model fusion is proposed.A model based on this algorithm model is built and simulation experiments are performed that.Finally,simulation results show that the performance of the proposed failure location algorithm model based on multi-model fusion is better than the single failure location algorithm model.