?-OTDR Pattern Recognition Based On Expert System

The distributed optical fiber sensing system based on the phase sensitive time domain reflectometer(?-OTDR)has the characteristics of simple and stable structure,high spatial resolution,and multi-point simultaneous monitoring.It is widely used in military,nuclear industry,petrochemical,medical and other fields,and has become a research hotspot in the field of distributed optical fiber sensing.Aiming at the ?-OTDR pattern recognition method,this paper designs the experimental data collection,synthesizes the respective advantages of different classification principles,and puts forward the ?-OTDR pattern recognition method based on expert system,which can effectively improve the event classification accuracy of ?-OTDR system.The main achievements of this dissertation are as follows:(1)Based on the theoretical research of ?-OTDR,an experimental device is built to collect the time-domain signals of four kinds of events: trampling,jumping,knocking and nondisturbance.(2)The signal preprocessing in time domain is completed,and the target of location is achieved by the method of feature band energy extraction.(3)The method of empirical mode decomposition and reconstruction combined with spectral subtraction is proposed to further reduce the noise of the above signals.The average signal-to-noise ratio after noise reduction is 17.88,and it is divided into 32 trampling events,30 jumping events,30 knocking events,50 undisturbed events,a total of 142 samples.(4)Using the intra class distance and inter class distance as the evaluation criteria,12 dimensional eigenvectors are extracted from the time-domain signal after noise reduction,which are respectively: maximum,minimum,average,rectification average,waveform factor,pulse factor,peak factor and margin factor and the first four IMF energy of each sample;after noise reduction signal division,the first 1000 data of each sample are taken as 1-D CNN sample data.(5)An expert system classification structure based on the model is proposed.Three sub classifiers and the model are trained and adjusted through 5 fold cross validation.The average validation classification accuracy is SVM: 91.70%,BP neural network: 92.56%,1-D CNN: 96.42%,and the model-based expert system classification structure is 98.75%.The classification method proposed in this dissertation is better than that of single classifier,which verifies the effectiveness of the model-based expert system classification structure for the improvement of classification accuracy.