Research On Fault Diagnosis Method Of Pumping Well Based On Convolutional Neural Network

Oil is one of the most widely used energy sources in industrial production and daily life.It occupies an important position in the energy field.As a kind of natural energy,the pumping wells used for petroleum exploitation are in bad working environment and often fail,which may affect the oil output at least or cause serious casualties and property losses at worst.Therefore,the safe and efficient operation of the pumping well is of great significance to the oil exploitation,and the key to the safety of the oil well is the accurate identification of the fault of the pumping unit.In the fault identification of the pumping unit,the basis of the dynamometer diagram is the commonly used method at present.Most of the existing methods use fault classification after extracting features from dynamometer diagrams as the main approach.The recognition accuracy of such methods has a strong dependence on feature extraction.Convolutional neural network has been widely used in the field of fault diagnosis due to its advantages of feature extraction and classification.It is of great significance to use convolutional neural network for fault diagnosis of pumping unit.This thesis takes the most widely used rod-type pumping unit as the research object.Based on the actual data collected from oil wells in Liaohe oilfield,the WIA-PA wireless dynamometer is used on site to collect data and draw the well-dynamometer diagram.Aiming at the difference between the uphole dynamometer diagram and the downhole pump dynamometer in reflecting the downhole part of the pumping unit,based on the Gibbs wave equation,the uphole dynamometer diagram was converted into a downhole pump diagram.Data preprocessing and data set augmentation are used to construct WPD oil well fault diagnosis data set.In the selection of convolutional neural network types,compare a variety of common convolutional neural network models,select the GoogLeNet network structure for pumping unit fault diagnosis.The activation function,normalization layer,full connection layer,learning rate,dropout layer and other important parameters of GoogLeNet network structure are modified to improve the network.The experimental results show that the improved convolutional neural network New-GoogLeNet can improve both the accuracy and speed of fault identification of the pumping unit.Further,in view of the impact of insufficient data volume on the identification accuracy of the existing oil pump fault data set,transfer learning is introduced to improve the identification accuracy.In the implementation of transfer learning,comparative analysis are conducted on migration of different data sets,different data set expansion methods and different freezing methods.The experimental results show that the improved convolutional neural network TLNew-GoogLeNet introduced with transfer learning further improves the fault recognition accuracy and recognition speed.In this thesis,the convolutional neural network is used to identify the faults of the pumping unit.The results show that it has the advantages of high accuracy and high recognition speed.The research method has important application value for the fault diagnosis of the pumping unit.