Research On Signal Pattern Recognition Method Of Sagnac Distributed Optical Fiber Sensing System

Distributed optical fiber sensor takes optical fiber as sensing element and transmission medium of sensing signal.It can realize long-distance and wide-range sensing.It can also achieve high-quality sensing at low cost in complex environment such as lack of communication and power.It has irreplaceable advantages for other sensors.However,due to the problems of long detection distance,laser source noise and external environment interference,distributed optical fiber sensing signals are accompanied by a large number of non-intrusive signals,which leads to high false alarm rate of the system.Therefore,it is one of the key points of the system to separate intrusive signals from the sensing signals accompanying non-intrusive signals.An effective intrusion signal extraction algorithm can accurately and quickly separate the intrusive signals from the sensing signals accompanying non-intrusive signals.The intrusion signal reduces the amount of calculation when the system identifies the intrusion signal in the later stage,and improves the recognition rate of the intrusion signal in the later stage of the system.The identification of intrusion sensor signal is another key point of the system,which refers to the identification of the types of intrusion sensor signals.Fiber optic sensing system responds to the phase modulation of optical signals.The waveforms of these sensing signals are often very similar,which increases the difficulty of classification and the false alarm rate of the system.In recent years,how to characterize the intrusion signal of optical fibers with comprehensive features and how to select a suitable classifier to classify the intrusion signal are also the research difficulties.In this paper,distributed optical fiber sensing is studied in many aspects.The main contents are as follows:1.Analyzing the basic sensing principle of optical fiber as a sensing element,studying the sensing principle of Sagnac type and several other commonly used distributed optical fiber sensing structures,and summarizing the current situation of intrusion signal pattern recognition of Sagnac type distributed optical fiber sensing system.2.In view of the inaccurate extraction of distributed optical fiber intrusion signals,which often results in the separation of intrusion signals containing redundant signals,the computational load of feature extraction and pattern recognition is increased,and the recognition rate of intrusion signals is reduced.A LC(Level Crossing,LC)algorithm combining short-time overvoltage and short-time logarithmic energy is proposed to extract intrusion signals and calculate adaptive threshold of CA-CFAR.Algorithm.3.In the aspect of distributed optical fiber vibration signal extraction,wavelet transform and EMD decomposition are used to extract the frequency domain features of distributed optical fiber vibration signal,such as kurtosis,energy,energy proportion,etc.Combining with traditional time domain features,a multi-dimensional joint feature is formed to characterize the optical fiber intrusion signal.4.Construct Support Vector Machine(SVM)and Extreme Learning Machine(ELM)to recognize intrusion signals,and compare the recognition rate and stability of different combination of intrusion signal features in different classification algorithms.