| With the rapid development of optical fiber communication,the research of optical fiber sensing is in the ascendant.Using optical fiber as a sensor device can sense the measured signal and transmit information simultaneously.It also has significant advantages such as good electrical insulation,corrosion resistance,electromagnetic interference resistance,and long sensing distance.In particular,Brillouin distributed optical fiber sensing technology,due to its high-performance temperature and strain sensing,is used widely in various industries.At present,for Brillouin optical fiber sensing,there are several problems such as the measurement accuracy and spatial resolution are not very ideal,the measurement time is long,and the double-peak spectrum is identified inaccurately.The key to solving these problems is to improve the methods and techniques of measurement of Brillouin frequency shift(BFS).Based on the design and development project of Brillouin Optical Time Domain Reflector(BOTDR)prototype for a company in Tianjin,and aimed to meet the technical requirements for high measurement accuracy,low measurement time,and accurate double-peak spectrum identification of distributed fiber optic sensing systems in engineering applications.In this paper,the coherent measurement technology and data processing methods of fiber optic BFS are studied deeply.Design and build a Brillouin distributed optical fiber sensing experimental system,in which the heterodyne method is used for coherent detection to obtain Brillouin scattering signal data.Based on the demodulation method of short-time Fourier transform(STFT),the noise reduction principle of frequency domain cumulative average is theoretically studied,and a control method for BFS measurement accuracy is proposed to adaptively select cumulative average times,which can maximize BFS measurement efficiency.Derived and analyzed the least squares optimal solution for Lorentz line fitting,proposed a main peak interception algorithm based on support vector machines(SVM)and a data filtering algorithm based on half-peak fitting,both of which can effectively improve the fitting effect and enhance the accuracy of BFS measurement.For the BFS error caused by the doublepeak spectrum,a single-step sliding window is used for fine analysis of the double-peak spectrum and three algorithms are proposed.They are doublepeak identification based on SVM,spectrum peak division based on improved K-Nearest Neighbors(KNN),and spectrum peak fitting based on half-peak fitting.All of these can effectively improve the BFS measurement accuracy in temperature change or strain transition section,and increase the positioning accuracy of the section to the limit determined by the sampling rate.The main work is as follows:To enhance the accuracy of BFS,a main peak truncation algorithm based on SVM is proposed.The SVM model is used to classify the Brillouin power spectrum data,and then the main peak data is corrected and fitted based on the linear fitting of noise data.Then more accurate Lorentzian curve is obtained.Experiment results show that the standard deviation of BFS is improved from 20.54 MHz to 1.81 MHz confirming that the algorithm can significantly improve the stability of BFS measurement.And the average measurement error of temperature is about 0.8℃,which meets the requirements of engineering applications.To reduce the measurement time of BFS,based on the characteristics of the least squares optimal solution of Lorentz fitting,a data filtering algorithm based on half peak fitting is proposed,which selects the upper half peak data of the power spectrum for curve fitting,eliminates the interference of data with a low signal-to-noise ratio.The algorithm selects data quickly,significantly reduces the amount of fitting data,and effectively reduces the computational complexity.It can reduce approximately 80%of fitting operations and effectively reduce measurement time.The experiment results show that the measurement accuracy of BFS obtained by half peak fitting is 1.56 MHz,which is improved by-35%compared with particle swarm optimization and Levenberg-Marquardt iterative algorithms,and the fitting error is reduced by 58.5%.The algorithm can significantly reduce computational complexity and improve real-time measurement while ensuring measurement accuracy.To fit the double-peak spectrum effectively,three data processing algorithms are proposed.The improved KNN spectrum peak division algorithm uses the first and last groups of power spectra in the temperature change or strain events transition section,classifies the double-peak spectrum data of the section to obtain two stable BFS curves,determines BFS distribution based on peak comparison.The double-peak identification algorithm based on SVM uses Brillouin scattering power data to construct a training set that includes four-dimensional input feature vectors such as frequency,normalized power,a first-order difference of power,and a second-order difference of power.The double-peak spectrum data is classified into two categories,and the minor peaks are cleared as noise data.The proposed half-peak fitting algorithm combined with the determination of data monotony can clear up the minor peak data during data filtering,and only select the upper half peak of the main peak for curve fitting.All three algorithms can improve the measurement accuracy of temperature change or strain events transition section,and make the spatial resolution of jump location reach the limit determined by the sampling rate,which is 0.1 m at a sampling rate of 1 GHz.In addition,to improve the efficiency of BFS measurement,the distribution characteristics of Brillouin scattering power spectrum data are statistically analyzed,and the statistic correlation between the cumulative average number and the power value error,as well as the transfer function between the power spectrum error and the BFS error,are derived.A control method for BFS measurement accuracy based on the power spectrum cumulative average is proposed.Experiments show that the system can automatically select an appropriate cumulative average number according to the accuracy requirements,and the control deviation is less than 5%under various accuracy indicators.Confirmed that the method can optimize the BFS measurement efficiency. |
