Research On Localization And Pattern Recognition Method Of Disturbance Signal For Dual M-Z Distributed Optical Fiber Sensor

Optical fiber sensing technology has many advantages such as long distance measurement,high measurement accuracy,and resistance to electromagnetic interference.It has become an important research direction in the field of perimeter protection.The distributed fiber perimeter protection system can be divided into: interferometric sensing technology and backscatter optical sensing technology.Sensing system with dual Mach-Zehnder interference technology in distributed fiber perimeter protection system,which has the advantages of sensitive sensing,simple positioning and demodulation methods,low cost and other.However,the real-time and accuracy of the disturbance location technology and the classification accuracy of the sensing signal mixed mode are still problems in the system.In this thesis,the two problems in the dual M-Z interferometric fiber-optic sensor system are studied separately,and verified by theoretical simulation analysis and experimental methods.Firstly,the disturbance signal localization algorithms for dual M-Z distributed optical fiber sensors are studied,and the relationship between parameters such as sampling frequency and signal-to-noise ratio of disturbance signals and positioning accuracy is analyzed.The quadratic cross-correlation algorithm is applied to the estimation of the perturbation location of the intrusion signal,and the cross correlation algorithm is transformed to the frequency domain by the Wiener-Khintchine theorem to perform the conjugate product operation,and then the inverse signal is used to calculate the perturbation signal.The estimated root-mean-square error at the 6-km disturbance location is reduced by 140 m compared to the cross-correlation estimation method.At the same time,the average time of a single positioning calculation is 27 ms to ensure the real-time performance of the system.Then,the disturbance signals detected by the dual Mach-Zehnder distributed optical fiber sensors have different characteristics of vibration frequency and amplitude,and different types of disturbance signals are distinguished to reduce the false alarm rate of the system to identify the disturbance signal as an intrusion event.In this thesis,The time-frequency characteristics of the disturbance signal were extracted by using discrete wavelet transform,short-term over-threshold rate.At the same time,the BP classification method and the support vector machine method are used to compare the results of the pattern classification of the data.Discrete wavelet decomposition was used to extract the energy spectrum characteristics of the sensor signal in different frequency bands.The short-term over-threshold detection method was used to extract the signal over-threshold rate characteristics.The results of the two methods were applied to the SVM classification algorithm and a comparative analysis was performed.On the basis of this,the parameters of the support vector machine classification method were optimized using genetic algorithms.Finally,based on the dual M-Z distributed optical fiber sensing experiment platform,the pattern recognition algorithm was tested and verified.The results of the experiment show that the improved SVM classification algorithm can obtain higher accuracy classification results under the 400 training samples and 100 identification samples.