Research Of Fault Location In Optical Networks Based On Deep Learning

With the rapid development of optical networks information technology,the transmission rate of optical communication networks is getting higher and higher,and the optical communication network as a network infrastructure is increasingly important in the communication systems.Various new applications in the Internet rely on intelligent,transparent optical networks to meet their high-speed communication requirements.A failure in an optical communication network,even a short-term service interruption,will inevitably affect network transmission.It even leads to the loss of important data,which is a very serious problem for an information data era.It can be seen that the fast and accurate fault location technology in the optical network is the basis for ensuring the normal operation of a large number of network services and ensuring the complete and reliable transmission of service data.Therefore,fault location in optical networks is an extremely important research field.The increasing scaleand complexity of optical networks has made fault location in optical network a challenging research topic.According to the research status of fault location technology in optical networks,this thesis proposes an alarm data preprocessing strategy based on distributed FP-Growth algorithm.Based on this,based on convolutional neural networks and deep neural networks fault location methods are proposed.The specific research work is as follows.Firstly,in order to obtain a set of alarm data samples that can be used for model training,the thesis proposes an alarm data preprocessing strategy based on the distributed FP-Growth algorithm.The original alarm data is collected from the data collection interface in the optical network through the collection system.And,the original alarm data is processed for alarm reason translation,alarm format conversion.Then,based on the distributed FP-Growth algorithm,the association rule analysis between the alarm data is completed,the alarm data expert knowledge database is constructed,and the alarm data labeling work is completed.Finally,considering the problem that the original alarm data distribution is extremely uneven,an alarm data enhancement strategy based on the alarm data association rules is proposed.A reliable and complete alarm data sample set is obtained,which lays a foundation for the later model training work.The results show that the running time of alarm data preprocessing based on distributed FP-Growth algorithm is short,between 0.5?3.5s.Secondly,in order to realize accurate and rapid fault location in optical communication networks,the thesis studies the convolutional neural networks and proposes fault location based on convolutional neural networks.First of all,the pre-processed alarm data is upgraded in dimension to obtain the alarm sample set that meets the requirements.Then,the fault location is modeled based on the convolutional neural networks,and the fault location model structure is determined according to the longitudinal indicators of the model.Finally,the fault location model is used to locate the fault and compare it with the existing fault location algorithm.The results fully prove that fault location model based on convolutional neural network can locate faults in the optical network more timely and accurately.The fault location delay is between 0.30-0.35ms,and the fault location accuracy is 90%and above.Thirdly,in order to make full use of the potential features of alarm data,and further improve the performance of fault location in optical communication networks,the thesis studies the deep neural networks and proposes fault location based on deep neural networks.Above all,the structure and principle of deep neural networks are deeply analyzed.Then,the fault location model of different parameters is trained and verified by using the pre-processed alarm data sample set.The parameters of the optimal fault location model based on deep neural networks are determined.Finally,comparing with the fault location model based on convolutional neural networks proposed above and the existing fault location algorithm in many aspects,the experimental results show that fault location based on deep neural networks has higher accuracy,reaching 95%and above,and the fault location delay is between 0.30?0.40ms.