Research On Fault Diagnosis Method Of Sucker Rod Pumping Wells Based On Generative Adversarial Networks

Currently,sucker rod pumping system is the main mechanical oil extraction method in domestic oil fields.Most of China’s oilfields developed earlier,and there is no perfect mining concept for oilfields.At the time,China’s industrial level was backward and there was no perfect oil extraction technology.This kind of oilfield exploitation environment has caused the oil well efficiency of the old oilfields in China to be lower and lower,and the oil wells have failed.The dynamometer card contains the actual working information of the oil well and is an important theoretical basis for the current use of artificial diagnosis in oil fields.With the continuous development of intelligent diagnostic technology,automatic diagnosis of oil well conditions based on dynamometer card has become mainstream.Based on the collection and theoretical analysis of dynamometer card,this paper takes intelligent diagnostic methods as a tool to establish a fault diagnosis model for sucker rod pumping system.Firstly,this paper introduces in detail the basic principle and working process of the oil extraction with rod pump,and based on this,the formation reasons and graphic characteristics of theoretical dynamometer card and the seven typical working condition dynamometer card collected at the site are analyzed.Secondly,due to the rough production of oil fields,it is difficult to collect on-site dynamometer card.The lack of dynamometer card makes the oil well fault diagnosis model less accurate.Therefore,this paper uses conditional generative adversarial networks to derive the sample of the dynamometer card to achieve the purpose of data enhancement.Through the generate effect of derived dynamometer card and improving the accuracy of fault diagnosis can prove the validity of the derived model.Thirdly,this paper use Zernike moment to extract the eigenvectors of the dynamometer card.This paper use probabilistic neural network with single smoothing factor and multiple smoothing factors to establish the classification model of oil well conditions.Manually determining probabilistic neural network parameters of smoothing factors is difficult.So using cuckoo search algorithm to optimize the parameters of the diagnostic model,which has a higher accuracy.Finally,we developed oil well monitoring equipment and used it to diagnose oil well conditions.From the diagnosis results,we can know that the equipment can effectively diagnose oil well conditions and meet the needs of oilfield production.