High Precision Recognition Method Of DAS Signal Of Communication Cables Based On Model Fusion

With the increasing demand of 5G,Internet of things,industrial Internet and data center in China,optical fiber and cable,as the foundation of communication network construction,has become the information artery of national economic development,providing good physical conditions for large-scale environmental perception and detection.The DAS system based on ?-OTDR technology,which can realize longdistance monitoring,can meet the needs of communication cable safety monitoring,and can realize long-distance,large-scale and all-weather intelligent security monitoring of communication cable.However,the classification and identification of interference events in DAS based communication optical cable security monitoring system is still facing many challenges.Based on this,this thesis starts from the time domain,frequency domain,time-frequency domain and model coefficient method,combines with ANN algorithm to extract the features of DAS signal of typical interference events of communication optical cable,which is more comprehensive than using the simple timefrequency domain features.In terms of classification algorithm,this thesis proposes a DAS signal classification and recognition model based on model fusion,which requires less data samples than the traditional recognition algorithm using complex deep learning model.The work of this thesis is as follows :(1)Based on the time and space characteristics of the signal,the feature extraction method of the interference event of the communication optical cable is studied.The characteristic parameters of the signal in the time domain,frequency domain,time frequency domain and model parameters are extracted.The multi-dimensional extraction of the signal characterization parameters can describe the signal characteristics better.In order to make up for the shortage of artificial extracted features,the deep learning method is used to further mine and classify the interference signal characteristics.And the ANN network structure is optimized from the network layer,activation function and optimization algorithm to improve the performance of the model.(2)The stacking algorithm is introduced to the DAS signal recognition of communication optical cable for the first time to fuse the traditional machine learning model.The test is carried out on the typical interference event data set of communication optical cable collected on the spot.The accuracy of the algorithm proposed in this thesis reaches 99.07%,which is 2.77% higher than that of the traditional machine learning model XGB,which is the best performance.The results show the effectiveness of the model fusion method.(3)This thesis constructs different reference models,analyzes and compares the DAS signal recognition effect of the proposed algorithm and the reference model for interference events near the communication optical cable from different classification and evaluation indexes.The experimental results show that the DAS signal classification and recognition method based on model fusion proposed in this thesis is better than other classification models for four typical interference events in the complex background environment of communication cable.Furthermore,the effectiveness of the proposed method is verified.From the point of view of real-time and stability,it is verified that the algorithm proposed in this thesis is suitable for the practical application of communication cable security monitoring based on DAS system.