Recognition Of Intrusion Events In The Periphery Of Oil And Gas Pipelines Under Complex Environment Based On Transfer Learning

As a fuel and chemical raw material,oil and natural gas play an extremely important role in production and life.Pipeline transportation is the most important method for long-distance transportation of oil and gas resources.Due to the danger of pipeline damage,the research on the identification of perimeter intrusion events has been widely concerned by all walks of life.The operation environment of oil and gas pipelines is complicated along the way,and the premise that the standard samples assumed in the traditional method are consistent with the actual samples is destroyed,resulting in a single recognition model under different environments to reduce the accuracy of intrusion event recognition.In this regard,this paper introduces the transfer learning method into the field of oil and gas pipeline security to provide new ideas for the identification of intrusion events.The main research contents are as follows:(1)First of all,in-depth understanding of the identification principle of intrusion events in the periphery of oil and gas pipelines,from the three perspectives of original signal,feature model,and recognition model,analyze the difference of intrusion events and interference event signals caused by complex environments,and further elaborate the research of this paper value.(2)For the problem of scarce samples in the target domain and missing labels,study the semi-supervised transfer learning method based on deep confrontation,use the deep convolutional neural network feature extractor F to replace the concept of generating the confrontation network generator,and automatically obtain the source and target domain features Space;the domain discriminator D solves the domain adaptation problem between the source domain and the target domain;under the action of the classifier G,it realizes the identification of intrusion events around the periphery of oil and gas pipelines.Through a large number of comparison experiments,the key model influence factors such as the number of convolution layers and the number of convolution kernels are analyzed.From the perspective of model recognition rate and field application effect,the effectiveness of the method in this paper is verified.(3)Aiming at the requirements of the scarcity of samples and high recognition accuracy in some areas of oil and gas pipelines,the supervised transfer learning model based on deep adaptation is studied,and the maximum mean difference is used to achieve the domain distribution difference measurement;the common maximum mean difference measurement method is used to face the source and target domains Different conditional distributions cause category confusion,difficulty in selecting kernel functions,and risk of single-layer measurement effects.A method for measuring the difference in distribution between edge distribution and conditional distribution,multi-angle kernel function observation,and multi-level feature coordination is proposed,combined with cross-entropy function.Realize knowledge transfer and intrusion event recognition.Through a lot of experiments,the key impact factors are analyzed,and the effectiveness of the method in this paper is proved from the perspective of model recognition rate and field application effect.