Research On Distributed Fiber Optic Sensing System Based On ?-OTDR And Target Classification Method

With the rapid development of science and technology,sensors are just like the nerves of human exploration and perception of the world,and the in-depth research on their technology has become particularly important.Distributed fiber optic sensing technology as a new field of sensor technology is also rising rapidly.The distributed Optical fiber sensing system based on phase-sensitive Optical Time Domain Reflectometer(?-OTDR)has the advantages of high positioning accuracy,low cost,multi-point positioning,etc.It has become one of the hot spots in the current research of Optical fiber sensing technology and has been widely applied in pipeline monitoring,border security,earthquake early warning,security alarm and other fields.In this paper,aiming at the problem of misjudgment of disturbance target recognition in the distributed fiber-optic sensing system of ?-OTDR,a probabilistic neural network algorithm(PNN)based disturbance event pattern recognition method is proposed to describe the target signal through multi-characteristic parameters,which effectively realizes the classification of different types of disturbance targets.Part of the experimental platform of ?-OTDR distributed optical fiber sensing system is designed.In the aspect of system design and characteristics of the signal preprocessing: first of all to ?-OTDR distributed optical fiber sensing system and its signal pattern recognition as the main research direction,from the analysis of the light scattering to study the ?-OTDR sensing technology and the simple theoretical research of the output signal,and making use of the MATLAB simulation of the ?-OTDR differential positioning simulation output signal and the airspace.The experimental platform of ?-OTDR distributed fiber optic sensing system was built,and the APD photo detector was designed.The pulse shaping module was designed to analyze the jitter problem of FPGA pulse signal.Of five kinds of actual situation: dig stolen,car pressure,hitting,trample,no disturbance,disturbance data collection and analysis,through the data normalization,mobile difference of output signal processing,and put forward the main wave energy(MVE),signals main lobe time delay(MVBW),the main peak(MDV),signal main wave Q(MVQV),signal spectrum(SFS)seven characteristic parameters as the signal characteristic value.In the aspect of characteristic signal classification recognition algorithm: take classification recognition probability neural network(PNN algorithm model,and the probability of PNN neural network,BP neural network,GRNN,comparing the generalized regression neural network(PNN in average operation time,identification accuracy and model accuracy stability all have certain advantages,it is concluded that the probability of PNN obvious advantages of neural network in target classification and recognition,on the PNN recognition test,three monitoring stations of percussion,the car,digging steal,trample,undisturbed five disturbance model experiment,the average recognition rate is respectively:91.21%,92.50%,90.96%,95.88%,the average recognition rate is above 90%.In view of the influence of smoothing factor value in PNN on network performance,a simulated annealing algorithm(SA)was proposed to optimize the smoothing factor,and the ability of SA algorithm to optimize the whole network was verified by mathematical model experiments.The optimized SA-PNN algorithm was used to conduct the field experiment,and the average recognition rate of the five disturbance modes of tapping,passing,digging,stealing,trampling and undisturbed at the three monitoring points was 95.02%,97.33%,93.60% and 99.12%,respectively.The accuracy rate was better than that of PNN algorithm.SA-PNN algorithm is more stable than PNN algorithm in accuracy stability.