Research On Effective Retrieval Method Of Fiber Brillouin Frequency Shift With Non-local Effects

Brillouin Optical Time-Domain Analysis(BOTDA)is a cutting-edge distributed optical fiber sensing technology at home and abroad.It has been successfully applied to temperature and strain measurement,structural health monitoring,geotechnical engineering,and leak detection along pipelines.Compared with other distributed optical fiber sensing technologies,BOTDA has the advantages of long sensing distance and high measurement accuracy.Non-local effects(NLE)is one of the important factors to limit the BOTDA sensing distance,which will cause the distortion of the measured Brillouin gain spectrum(BGS)and lead to serious measurement errors of Brillouin frequency shift(BFS).To solve this problem,this thesis mainly studies the effective retrieval method of fiber Brillouin frequency shift with non-local effects.The specific research contents of this thesis are as follows:(1)This thesis describes the basic principles of non-local effects in BOTDA.Based on the theoretical model of non-local effects,we focused on the analysis of the BGS asymmetric distribution,linewidth variation and the corresponding measurement errors caused by the non-local effects in the single-sideband BOTDA loss spectrum structure.(2)This thesis applies machine learning to BOTDA,and elaborates the specific process of constructing a neural network,which are suitable for processing BOTDA experimental data.It is proposed an add-noise processing on the datasets to avoid overfitting and effectively improve the adaptability of the network.(3)In this thesis,a 25-km experimental device based on single-sideband BOTDA loss spectrum measurement is designed.Through the multi-directional comparisons of experimental and simulated results with and without non-local effects,the consistency of theory and experiment is well verified.(4)In this thesis,the selections of samples in two kinds of neural networks datasets are explained in detail.By testing the BOTDA simulated data,the reliabilities of two neural networks for processing BOTDA experimental data are well verified.(5)Aiming at the deficiency of the poor accuracy of the Brillouin frequency shift at the disturbed region retrieved by the basic neural network(B-ANN)with non-local effects,this thesis proposes a specific neural network(NLE-ANN)to solve this problem.In order to further optimize the processing results,a cooperative implementation of B-ANN and NLE-ANN is proposed,which realizes the effective retrieval of Brillouin frequency shift along the whole fiber with non-local effects,and at the same time obtains lower measurement uncertainty.In the first four experimental cases,the root mean square errors(RMSEs)of NLE-ANN in the disturbed region are 1.26 MHz,1.45 MHz,1.51 MHz and 3.58 MHz,while the RMSEs of B-ANN(LCF)are 6.71(6.89)MHz,7.82(8.11)MHz,11.13(11.40)MHz and 14.41(14.75)MHz,respectively.Therefore,in terms of root mean square errors,NLE-ANN is obviously better than B-ANN and LCF.