基于Laplacian正则和协同训练的抽油井故障诊断方法  

Fault Diagnosis Method for Sucker Rod Pumping Wells Based on Laplacian Regularization and Co-training

作　　者：杨浩然 袁春华 Yang Haoran;Yuan Chunhua(School of Automation and Electrical Engineering,Shenyang University of Technology,Shenyang 110159,China)

机构地区：[1]沈阳理工大学自动化与电气工程学院,沈阳110159

出　　处：《机电工程技术》2024年第6期216-220,共5页Mechanical & Electrical Engineering Technology

基　　金：辽宁省本科教改优质教学资源建设与共享项目(SBKJGYZ-2021-06)。

摘　　要：单一视角下部分不同类型的工况样本相似会导致故障诊断准确率不高、无标记样本未被使用等问题,为进一步提高有杆泵抽油井故障诊断准确率,有效利用无标记样本,充分利用示功图和电功率两种数据,采用Laplacian正则和协同训练算法设计了抽油井故障诊断方法。首先,将Laplacian正则引入到协同训练模型中,以提高初始分类器精度;其次,以示功图和电功率作为特征视角,利用三层小波包变换提取视角特征,使用特征数据训练出对应的初始分类器,两者之间交换未知信息并更新分类器,提高模型的精确性;最后,应用该方法对某采油厂抽油机井的6种典型工况进行识别,并与标准协同训练、有监督学习方法进行比较。结果表明:在有标记样本较少时,该方法识别准确率优于仅使用示功图的SVM、仅使用电功率的SVM、标准协同训练,分别提高了约23.4%、20.83%、3.42%,验证了该方法的有效性。Similar samples of different types of working conditions from a single perspective can lead to low accuracy in fault diagnosis and the use of unlabeled samples.In order to further improve the fault diagnosis accuracy of sucker rod pumping wells,effectively utilizing unlabeled samples,and make full use of dynamometer cards and motor power,a fault diagnosis method for pumping wells is designed by using Laplacian regularization and co-training algorithms.Firstly,the Laplacian regularization is introduced into the co-training model to improve the accuracy of the initial classifier.Secondly,dynamometer cards and motor power are used as the feature perspectives,the three-layer wavelet packet transform is used to extract the perspective features,and the corresponding initial classifier is trained by using the feature data.The unknown information is exchanged and the classifier is updated to improve the accuracy of the model.Finally,the method has been applied to identify six typical operating conditions of a pumping well in a certain oil production plant,and compared with standard co-training algorithm and supervised learning methods.The results show that when there are fewer labeled samples,the recognition accuracy of the proposed method is better than that of SVM using dynamometer cards only,SVM using motor power only,and standard co-training,which are improved by about 23.4%,20.83%and 3.42%,respectively,which verifies the effectiveness of the proposed method.

关 键 词：故障诊断 协同训练 Laplacian正则 抽油井 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]