Research On Recognition Algorithm Of Optical Fiber Vibration Signals Based On Ensemble Learning

The optical fiber pre-warning system can locate the vibration and recognize the vibration type by collecting and analyzing the vibration signal in the distributed optical fiber sensors buried underground.And the recognition of optical fiber vibration signal is of high research value.At present,most of the signal recognition algorithms are analyzed based on the manual feature extraction.Although they can be used to complete the recognition task,they are time-consuming and laborious,and require a lot of prior knowledge in signal processing.Therefore,in this paper,the vibration signals that have been simply processed are taken as input and the neural networks are used as classifiers for signal recognition.At the same time,due to the limited recognition effect of the neural network model,this paper uses ensemble learning method to further improve the recognition effect of the network model.Firstly,the paper introduces the optical fiber pre-warning system used to collect optical fiber vibration signals and analyzes the characteristics of the signals.Since the dimension of optical fiber vibration signals is high and the dataset of signals is small,if a large scale deep neural network is used,the training will be inadequate.So,a small stochastic configuration network(SCN)is used for signal recognition.In the research process,we found that although the use of stochastic configuration network can leave the work of extracting signal features to machine,and improve the recognition of signals.However,the performance of a small three-layer neural network has room for improvement.Therefore,this paper combines SCN with ensemble learning methods to further improve the recognition effect of neural network.First,the Ada Boost-SCN.v1,which uses SCN as base classifier,is combined with Ada Boost method,and the obtained model can effectively improve the recognition effect of optical fiber vibration signals.In the complex working environment,OFPS is susceptible to natural or artificial interference,resulting in the optical fiber vibration signal contains many kinds of noise.On the one hand,it is necessary to improve the recognition of the noisy vibration signals in the actual scene.On the other hand,using the noisy signals as the training set can improve the robustness of the model.In addition,although Ada Boost-SCN.v1 effectively improved the recognition effect of SCN,the method took longer training time,occupied more resources,and did not highlight more features of SCN.Therefore,according to the SCN's characteristics of generating hidden layer nodes one by one,the ideas of Ada Boost and Bagging are combined with the network,and three improved models,Bootstrap-SCN,Ada Boost-SCN.v2 and Ada Boost-Bootstrap-SCN,were proposed,and experimental verification of these models was carried out.As can be seen from the experimental results,the recognition effect of noiseless vibration signal increased from 86.1% of original SCN to 99.6% of Ada Boost-Bootstrap-SCN.At the same time,the average recognition effect of the vibration signal with four kinds of common noises increased from 58.7% of original SCN to 81.3% of Ada BoostBootstrap-SCN.