基于卷积神经网络模型的抽油机系统故障诊断  

Pumping system fault diagnosis based on convolutional neural network model

作　　者：刘芑辰 冯子明[1] 蒋国斌[2] 孙桐建[2] 李琦[2] LIU Qichen;FENG Ziming;JIANG Guobin;SUN Tongjian;LI Qi(School of Mechanical Science and Engineering,Northeast Petroleum University,Daqing 163318,Heilongjiang,China;Petroleum Production Engineering Research Institute,PetroChina Daqing Oilfield Company,Daqing 163000,Heilongjiang,China)

机构地区：[1]东北石油大学机械科学与工程学院 [2]中国石油大庆油田有限责任公司采油工程研究院

出　　处：《石油钻采工艺》2021年第6期777-781,共5页Oil Drilling & Production Technology

基　　金：国家自然科学基金项目“变速驱机-杆-泵全耦合动力学行为及优化运行节能机理研究”(编号:51774091);黑龙江省博士后启动基金项目(编号:LBH-Q20083);黑龙江省自然科学基金联合引导项目(编号:LH2019E018)。

摘　　要：陆地机械采油普遍采用有杆抽油系统,示功图是油井工况的重要指示。在实际开采过程中,由于抽油井数量大、分布广,人工检测油井耗时费力。为提高人工检修效率、提升自动化水平,针对示功图的图形特征,在卷积神经网络Le-Net模型的基础上,建立简化卷积神经网络模型。收集实际生产数据经预处理后输入机器学习模型进行训练,得到关于示功图的分类模型,同时通过测试集数据对分类结果进行评价。结果表明,建立的卷积神经网络模型具有良好的稳定性,能够通过数据学习得出准确率较高的分类模型;所建立的分类模型能够稳定处理多分类问题,对于15种故障类型分类实现效果良好;通过测试集进行评价,该模型准确率达92%以上,预测效果可以满足油田实际生产需求。At present,sucker rod pumping system is commonly used for onland mechanical oil production,and indicator diagram is an important indicator of oil well working condition.In the process of actual production,manual oil well detection is time consuming and laborious due to the large number and wide distribution of pumping wells.In order to improve manual maintenance efficiency and automatic level,a simplified convolutional neural network model was established based on the convolutional neural network Le-Net model,according to the graphic characteristics of indicator diagram.The actual production data were input the machine learning model for training after they were collected and pretreated,and thus the classification model of indicator diagram was established.In addition,the classification was evaluated by using the test set data.The results show that the newly established convolutional neural network model is stable and can provide the classification of higher accuracy by means of data learning.The classification model can treat multi-classification problems stably and does well in the classification of 15 types of faults.What’s more,the evaluation on the test set indicates that the accuracy of the model is up to 92%and its prediction effect can meet the demand of actual oilfield production.

关 键 词：抽油机 示功图 卷积神经网络 故障诊断 

分 类 号：TE355[石油与天然气工程—油气田开发工程]