Research On Pattern Recognition Method Of Distributed Optical Fiber Vibration Sensing System

Phase-Sensitive Optical Time Domain Reflectometer((37)OTDR-),The distributed optical fiber vibration sensing system adopts optical fiber as transducer and transmission,which has the advantages of simple structure,convenient use,wide measurement range,anti-interference,and shows an increasingly important role in the field of modern artificial intelligence measurement.As the continuous development of modern artificial intelligence,it is not enough to just measure and locate vibration signals in the actual application.In the vibration events,it is also desirable to distinguish the type of vibration signals and response to actual vibration events.Therefore,it is necessary to research the pattern recognition arithmetic of the phase sensitive vibration signal.This dissertation mainly focuses on the pa ttern recognition arithmetic of distributed optical fiber sensing system,including the feature acquisition and classification for vibration signals,comparison and improvement on different algorithms to increase accuracy and reduce false positive rate and leakage.The main content of this dissertation is as follows:1.This paper introduces the distributed optical fiber vibration sensing system,including the state of art in the field of pattern recognition,structures of the arithmetic,and theoretical construction of the OTDR model.2.In this paper,the signal processing for vibration detection is researched,it includes demodulation,localization,feature extraction and feature engineering.According to the characteristics of vibration signal,a feature extraction method is proposed,it combines Hilbert with modified ensemble empirical mode decomposition.The simulation and experiments are respectively carried out.Comparing the empirical mode decomposition and the complementary ensemble empirical mode decomposition method,it can be concluded that the MEEMD decomposition method has characteristics in high accuracy and minimum reconstruction error.Compared with the CEEMD,the MEEMD saves 3.1363 s in signal feature extraction.3.In this paper,combined with time domain,energy domain and other methods,the MEEMD method is mainly applied on the vibration-free signal,human walk,vibration isolation network,heavy object fall and trolley passing events,etc.In the case of known vibration signal types,the experiment was repeated to obtain a set of eigenvectors of 3000 sets of vibration signals.4.In this paper,based on the characteristics of distributed optical fiber sensing system,a two-level classifier design is adopted to distinguish between environmental disturbance and human interference.The support vector machine(SVM)classification method is used for primary classification.According to the machine learning algorithm in artificial intelligence,the integrated learning algorithm is adopted as the second-level classifier,and two kinds of classifiers,random forest and Gradient Boosting Decision Tree(GBDT),are designed respectively to measure the accuracy of pattern recognition,it can be flexibly selected for different application scenarios.5.In this paper,80% of the extracted 3000 sets of feature vector groups are used as training sample data,and 20% is used as test sample data.The classifier parameters are tuned by cross-validation.In the same data sample,the accuracy of the two classifiers both reached 97% or more.