基于CNN的抽油机井故障诊断方法研究  

Research on Diagnostic Method of Oil Pumping Well Fault Based on Convolution Neural Network

作　　者：杜红勇[1] 隋顾磊 DU Hong-yong;SUI Gu-lei(Sinopec Dalian Research Institute of Petroleum and Petrochemicals,Dalian Liaoning 116045,China)

机构地区：[1]中国石化大连石油化工研究院,大连116045

出　　处：《当代化工》2021年第12期2909-2913,共5页Contemporary Chemical Industry

基　　金：中国石化集团公司总部科技项目(项目编号:320117)。

摘　　要：抽油机井故障判断的有效手段是对其示功图进行检测和识别。为了提高抽油机井故障判断的准确率和抽油机井生产管理的智能化水平,通过采集抽油机井示功图数据,建立了典型示功图样本库,并基于深度学习算法构建了示功图自动识别的卷积神经网络(CNN)模型。利用示功图样本库数据对CNN模型进行了训练和测试。结果表明:CNN模型收敛性好,模型训练准确率达到了99%,测试准确率达到了97%以上,达到了预期效果。该方法可用于抽油机井示功图分类和故障诊断。An effective method for judging the failure of a pumping unit well is to detect and identify its indicator diagram.In order to improve the accuracy of oil well fault diagnosis and the intelligent level of oil pumping well production management,the typical indicator diagram sample library was established by collecting the indicator diagram data of oil pumping wells.And based on deep learning algorithm,the convolutional neural network(CNN)automatic recognition model of indicator diagram was constructed.The CNN model was trained and tested by using indicator diagram sample library data.The results showed that the convergence of CNN model was well,the accuracy of model training was 99%,and the accuracy of testing was more than 97%,results reached the desired effect.The method could be used in the classify of oil pumping well indicator diagram and the diagnosis of oil pumping well fault.

关 键 词：深度学习 卷积神经网络 抽油机井 示功图 故障诊断 

分 类 号：TE938[石油与天然气工程—石油机械设备]