Research On Oil Pipeline Intrusion Detection Based On Deep Learning

Phase-Sensitive Optical Time Domain Reflectometer(Φ-OTDR)is a representative distributed fiber sensing technology,which is widely used in structural flaw detection,pipeline monitoring,homeland security and other vibration sensing fields.Based on the monitoring of oil pipeline intrusion events,this paper studies the signal processing and signal recognition of Φ-OTDR system.The principle of Φ-OTDR system applied in oil pipeline intrusion monitoring is studied,and some intrusion experiments are carried out at the same time.The most important part is the algorithm of location and recognition of intrusion events based on neural network.In order to solve the problem that the past algorithm is difficult to realize the event location and identification at the same time,a method based on the idea of target detection is proposed,and an improved target detection network is built to locate and identify three different kinds of intrusion events.The main achievements of this paper are shown as follows.1.The principle of Φ-OTDR system applied in oil pipeline intrusion monitoring is studied,and some intrusion experiments are carried out to get the target signal.The field experiments of human beating,human digging and human jumping are carried out on the surface of the pipeline to simulate the intrusion event and non-intrusion interference event,which lays a foundation for signal processing and signal analysis.2.The processing and analysis method of optical fiber signal are studied,especially the signal analysis method based on neural network.The process from signal acquisition to signal preprocessing and signal analysis is presented,and a new intrusion event monitoring method based on target detection network is proposed.In the experiment,YOLOv3(You Only Look Once Version 3)network is used as the basic network structure.The backbone and the loss function of classification are optimized to improve the accuracy and speed of the network.As a result,three different kinds of events can be located and identified precisely and quickly.3.The data augmentation of the sample is studied.Based on the objective requirements of data augmentation in the experiment,the influence of several traditional data augmentation methods on the network detection effect is discussed.In addition,the method based on generative adversarial networks is studied,and a deep convolutional generative adversarial network(DCGAN)is built to generate the new samples of three kinds of events.The experiment shows that the data augmentation method,including image shift and the method based on DCGAN,can effectively improve the detection effect of the network.