Research On Vibration Signal Recognition Of Distributed Optical Fiber Sensing System

The distributed optical fiber vibration sensing system is often used in the monitoring of safety systems due to its characteristics,such as the monitoring of potential safety hazards of oil lines,the monitoring of communication optical cable lines,the rail sensing along the railway and other fields.The phase change of Rayleigh scattering light,as the only condition of measuring system disturbance,is used in the phase sensitive optical time domain reflectometer(?-OTDR).It has the characteristics of simple structure,stable system,and multi-point monitoring at the same time.At present,?-OTDR is widely used in various practical application scenarios,with good development prospects and applicable space,so the research of ?-OTDR technology is also gradually developing.However,at present,there are still some limitations in the application of the pattern recognition method of optical fiber signal in the actual scene,such as the effect of denoising is not obvious,the feature extraction does not match,and the classification effect gap is large.In this thesis,by analyzing the types of disturbance events,preprocessing the initial data,using appropriate denoising methods,improving the feature extraction method and classification algorithm,forming a model recognition method based on moving average difference denoising,multi parameter feature extraction and classification fusion,this method has higher accuracy than single feature extraction.The main research work of this thesis is as follows:(1)For signal preprocessing and denoising,the optimization methods of correlation coefficient framing and moving average difference denoising are proposed.After the acquisition of the signal,first of all,preprocess the original signal,and propose a way of framing based on the correlation coefficient.By extracting the correlation coefficient between the vibration segments,extract and separate the vibration signal submerged in the environmental noise,effectively separating the vibration signal.On the basis of preprocessing,the signal is normalized and denoised by means of moving average difference.Combining the advantages of average method and difference method,the distortion points can be effectively reduced,the curve is smoother and the signal-to-noise ratio is increased by 9.48 d B.(2)Before feature extraction,signal features are analyzed and extracted.Considering the actual use scenario of optical fiber sensor,four kinds of disturbance signals are set,including knocking,rolling,climbing and simulating rain,so as to better restore the key disturbance of perimeter protection in the real scenario.Then the signal features are analyzed.Under the condition of removing the environmental noise,whether there is disturbance or not and whether the disturbance is artificial are discriminant analyzed.According to the analysis results,the disturbance features are extracted from different aspects to provide the feature vector for the subsequent classification.(3)A weighted classification method based on classification fusion theory is proposed for pattern recognition to enhance the accuracy of classification.After building a ?-OTDR sensor system,according to the data characteristics and signal characteristics,SVM(Support Vector Machine),BP(Back Propagation)and Adaboost(Adaptive Boosting)three different classifiers are used for classification,according to the characteristics of different types of disturbance signals,according to the way of weighted voting,the classifiers are fused,and then different types of disturbance are accurately classified,and the results of single classifier recognition are compared,and the average classification of classification fusion algorithm is achieved The accuracy is over 94%.