Research And Application Of Distributed Fiber Optic Vibration Event Identification Based On Time-space Signals Analysis

Distributed fiber optic vibration event recognition system is mainly composed of distributed sensing and vibration event recognition.It is widely used in surrounding safety,oil and gas pipeline monitoring,earthquake and tsunami disaster warning and other scenarios.The recognition accuracy,generalization ability,false-alarm and missing-alarm rate are the indicators to evaluate the goodness of this recognition system.In the research of fiber optic vibration event recognition technology,the machine-learning based recognition method has certain shortcomings,this kind of method attach great importance to the feature extraction method of vibration events and they often choose a variety of complex mixed feature extraction method.But too many feature extraction means need to consume a lot of time and cannot guarantee the real-time of the system.In this paper,we introduce unprocessed raw time-space domain signals with the richest feature information,and also introduce deep convolutional neural networks with powerful feature extraction capability,which are combined to propose an end-to-end vibration event recognition scheme that does not require manual feature extraction.The main research elements of this paper are as follows.(1)A distributed fiber optic vibration sensing event recognition system based on convolutional neural network was designed and the corresponding hardware system was built.A large number of vibration space-time domain signals of different typical events are collected by this system.(2)Using the feature extraction capability of convolutional neural networks,the core time-space features of different events are learned directly from the original time-space domain signals without human intervention.It was experimentally demonstrated that the method achieves an accuracy of more than 99% for the identification of three typical intrusion events and achieves fast and accurate end-to-end vibration event identification.(3)A comparative study of the proposed method with four other different recognition methods verifies that the proposed method has higher recognition accuracy,model generalization capability and practical potential.