The Research On Feature Extraction And Classification Algorithm Of Optical Fiber Vibration Signal Based On FDM Energy Entropy

The optical fiber pre-warning system(OFPS)is an pre-warning system that is applied around the pipeline using optical fiber sensing.It has the characteristics of good stability and high accuracy.It is widely used to detect pipeline leakage,man-made or mechanical intrusion.After OFPS detects the signal intrusion,the recognition part can use determine the type of intrusion signals by algorithm.At present,there have been many mature detection and recognition algorithms in the OFPS field.However,it is still a big challenge to accurately determine the type of the signal.This paper focuses on the feature extraction and recognition of fiber vibration signals.The proposed algorithm distinguishes the artificial signals from the mechanical signals.In order to improve the accuracy of the fiber vibration signal classification algorithm,this paper studies different adaptive time-frequency analysis methods and the improvement strategy,which finally realizes the classification of the four kinds of vibration signals and improves the efficiency of the algorithm.Firstly,this paper introduces two methods of single feature extraction based on time and frequency domain,and then proposes a sample entropy feature extraction method based on entropy domain according to the characteristics of different timing signals,and completes the feature extraction of two kinds of artificial intrusion signals,namely,manual tapping and digging signals.Secondly,in view of the weak adaptability of the single feature extraction method and the difficulty in identifying multi-type optical fiber vibration signals,this paper uses the time-frequency comprehensive analysis method to study it,and adopts the adaptive Variational Modal Decomposition(VMD)algorithm to process the signals.Aiming at the problem that VMD need to manually preset the modal decomposition parameters to get reasonable modal components,this paper puts forward the improved algorithm incorporating feedback model of VMD,which avoids the frame of manual parameter setting with improving the adaptability.We use the algorithm to extract marginal spectrum features of similar components of three signals,namely,manual tapping,pick planing,and electric drill.In addition,in view of the problem that the VMD algorithm based on the feedback mechanism is liable to lose information when reconstructing fiber vibration signal,we propose a feature extraction and classification algorithm based on FDM energy entropy.It proves that the classification effect is better and the adaptability is higher.The Fourier Decomposition Algorithm(FDM)algorithm uses a search mode from low frequency to high frequency(LTH)to automatically obtain the intrinsic model functions and count the total number.After that,we use the correlation theory to filter and reconstruct the FDM decomposition results.Then,we defines the FDM energy entropy feature based on the definition of information entropy and energy entropy Space theory,combining the FDM energy entropy of the fiber vibration signal with other important features into a feature vector.Finally,the algorithm realize the classification of the four fiber vibration signals of tapping,digging,running,and electric drill accurately combining with SVM.