Research On Abnormal Event Diagnosis Based On Phase-OTDR Vibration Signal

In recent years,distributed optical fiber sensing technology based on Phase-sensitive Optical Time Domain Reflectometer(Phase-OTDR)has been widely used in residential,Security monitoring in key areas such as traffic and military.In this paper,we study and design a machine learning algorithm based on the Phase-OTDR vibration signal to realize the identification and diagnosis of the above abnormal events by taking the possible human damage to the road surface,illegal construction and climbing road guardrails and other events that may occur on the traffic pavement.The specific research contents of this paper are as follows:(1)Preprocessing of Phase-OTDR vibration signal.According to the types of common abnormal events on traffic roads,three types of vibration signals generated by manual knocking,excavator digging and illegal climbing are selected as data sources.Aiming at the problems of modal aliasing and difficult signal reconstruction in the existing empirical mode decomposition algorithms,a new Phase-OTDR signal denoising method is proposed,which combines ensemble empirical mode decomposition and fast independent component analysis.Taking tapping signal as an example,the denoising effect of this method is compared with the existing wavelet method,ensemble empirical mode decomposition and fast independent component analysis method.The experimental results verify the effectiveness of this method.(2)Feature extraction of Phase-OTDR vibration signal.According to the characteristics of knocking,digging,climbing and noise signals after denoising,an extraction method of multi-dimensional features such as time domain,frequency domain and autocorrelation is proposed,and a random forest classification model is established to extract new features and The effectiveness of the method is verified by comparing the effects against the features.(3)Identification of Phase-OTDR vibration signals.Aiming at the characteristics of category imbalance and large amount of data existing in actual data,an integrated learning classification model based on Random Forest(RF),BP neural network and adaptive boosting(AdaBoost)is proposed to improve the classification performance of the algorithm.The method is compared with the single classifier classification method to verify the effectiveness of the algorithm.The machine learning algorithm based on the Phase-OTDR vibration signal proposed in this paper can better identify the knocking,digging and climbing signals,so as to detect the illegal construction,malicious damage and other abnormal events that may occur on the traffic road in time.And the monitoring and protection of the area has certain practical significance.