无杆泵井产液量虚拟计量方法对比研究  

Comparative study virtual measurement methods of liquid production in rodless pump wells

作　　者：冯钢 管桐[2] 燕云 李轩 王晓波 马丹 FENG Gang;GUAN Tong;YAN Yun;LI Xuan;WANG Xiaobo;MA Dan(Xi'an Supcon World Technology Development Co.,Ltd.,Xi'an Shaanxi 710016,China;Huabei Oilfield Engineering Technology Research Institute,Renqiu Hebei 062550,China;Associated Gas Comprehensive Utilization Project Department of PetroChina Changqing Oilfield Company,Xi'an Shaanxi 710000,China)

机构地区：[1]西安中控天地科技开发有限公司,陕西西安710016 [2]华北油田工程技术研究院,河北任丘062550 [3]中国石油长庆油田分公司伴生气综合利用项目部,陕西西安710000

出　　处：《石油化工应用》2023年第9期50-56,60,共8页Petrochemical Industry Application

摘　　要：针对现有无杆泵井计量以硬件设备为主,存在着占地面积大、建设投资费用高、人工巡井工作量大、单量数据少等问题,根据无杆泵井产液量多种虚拟计量方法优选一种基于XGBoost的大数据产液量动态计量方法。首先根据无杆泵井机采特征,消除不同井型、电机功率等影响因素,建立无因次关键参数,利用皮尔逊(Pearson)对数据样本进行了属性数据与产液量的关联性分析,并根据随机森林特征选择方法进行数据降维确定主控参数,定量研究无杆泵井生产数据变化规律与产液量之间的关系,选用XGBoost模型建立了一种广泛应用时序数据学习和预测的无杆泵井产液量预测模型。经过某区块10口无杆泵井实际产液量数据与IPR曲线拟合法、电参分析法进行对比分析,优选大数据产液量计算方法,产液量计算误差在±5%以内,解决了现场针对电潜泵、电潜螺杆泵等无杆采油方式无有效单井在线计量方法的难题,填补了技术空白。In response to the existing problems of large land area,high construction investment cost,large manual well inspection workload and limited single volume data,the measurement of rodless pump wells mainly relies on hardware equipment.Based on multiple virtual measurement methods for rodless pump wells'liquid production,a dynamic measurement method for big data liquid production based on XGBoost is selected.Firstly,according to the mechanical production characteristics of sucker rod pump wells,eliminate the influence factors such as different well types and motor power,establish dimensionless key parameters,use Pearson to analyze the correlation between attribute data and liquid production of data samples,and use random forest feature selection method to reduce data dimensions to determine the main control parameters,and quantitatively study the relationship between the production data change law and liquid production of sucker rod pump wells.The XGBoost model was selected to establish a widely used time series data learning and prediction model for predicting the liquid production of sucker rod pump wells.After comparing and analyzing the actual liquid production data of 10 rodless pump wells in a certain block with IPR curve fitting method and electric parameter analysis method,the big data liquid production calculation method was optimized,and the yield calculation error was within±5%,which solved the problem that there was no effective single well online measurement method for rodless oil production methods such as electric submersible pump and electric submersible screw pump,and filled the technical gap.

关 键 词：无杆泵 虚拟计量 电参分析法 IPR XGBoost 

分 类 号：TE355.5[石油与天然气工程—油气田开发工程]