基于稀疏自编码神经网络的抽油机井故障诊断

1.河南科技大学 机电工程学院,河南 洛阳 471003; 2.洛阳乾禾仪器有限公司,河南 洛阳 471000

稀疏自编码器; 示功图识别; 故障诊断; 深度学习; 无监督学习

Fault diagnosis of pumping well based on sparse auto-encoder neural network
ZHONG Zhi-dan1,FAN Hao-jie1,LI Peng-hui2

(1.School of Mechanical and Electrical Engineering,Henan University of Science and Technology,Luoyang 471003,China; 2.Luoyang Qianhe Instrument Company,Luoyang 471000,China)

sparse auto-encoder; indicator diagram recognition; fault diagnosis; deep learning; unsupervised learning

DOI: 10.13800/j.cnki.xakjdxxb.2018.0422

备注

针对传统示功图识别方法对抽油机井进行故障诊断存在人工选取示功图几何特征,识别准确度低等问题,提出一种基于稀疏自编码神经网络的示功图智能识别模型。采用无监督学习方式的稀疏自编码器构建特征学习网络自动提取无标签示功图训练集图像特征,然后采用有标签示功图训练集对softmax分类器进行有监督训练,最后通过稀疏自编码神经网络对学习到的有标签示功图测试集特征进行分类并给出故障诊断结果。结果 表明,将稀疏自编码神经网络应用于示功图识别,测试准确度能够达到99.44%,优于其它分类模型。稀疏自编码神经网络直接从像素层面提取所需要的特征,不需要人为选定设计特征,为提高示功图识别准确度提供了帮助,进而解决了抽油机井故障难以准确诊断的难题。

There are two main problems of thepumping well fault diagnosis by the traditional indicator diagram recognition methods:manual selection of indicator diagram's geometric characteristics and low recognition accuracy.An intelligent recognition model of indicator diagram based on sparse auto-encoder neural network is proposed.The sparse auto-encoder of unsupervised learning method is used to construct feature learning network to automatically extract the image features of the unlabeled indicator diagram training set.Then the softmax classifier is supervised training by the labeled indicator diagram training set.Finally,the learned features of the labeled indicator diagram test set are classified and the fault diagnosis results are given by sparse auto-encoder neural network.The experimental results show that when sparse auto-encoder neural network is applied to indicator diagram recognition,its accuracy reaches 99.44%,which is better than that of other classification models.The sparse auto-encoder neural network extracts the required features directly from the pixel level,and doesn't require the artificial features selection and design,which provides help to improve the recognition accuracy of indicator diagram and solves the problem of difficult to accurately diagnose the fault of pumping well.