基于SMOTE-DA-RF算法的有杆抽油系统井下工况识别  

Identification of underground operating condition of sucker rod pumping system based on SMOTE-DA-RF algorithm

作　　者：王通[1] 罗真伟 WANG Tong;LUO Zhen-wei(School of Electrical Engineering,Shenyang University of Technology,Shenyang 110870,China)

机构地区：[1]沈阳工业大学电气工程学院,沈阳110870

出　　处：《沈阳工业大学学报》2022年第1期84-89,共6页Journal of Shenyang University of Technology

基　　金：国家自然科学基金项目(61573088).

摘　　要：针对传统工况识别算法在识别有杆抽油系统工况时,存在生产措施调整滞后以及生产效率下降等问题,提出了一种基于改进的随机森林工况识别算法.采用灰度矩阵特征提取算法对泵功图进行特征提取,将灰度特征值通过合成少数类过采样技术进行上采样,实现不平衡数据均衡化;利用蜻蜓优化算法选取随机森林参数对抽油机井工况进行识别,并以辽河油田的生产数据进行实验验证.结果表明,该方法能够避免传统识别方法选取参考工况不准确的问题,减少不平衡数据对工况识别的影响,提高工况识别的准确率,能够满足油田现场的实际需求.In order to solve the problems of delayed adjustment of production measures and decreased production efficiency when a traditional identification algorithm of operating condition is used to recognize the operating conditions of sucker rod pumping system,an improved random forest(RF)recognition algorithm of operating condition was proposed.A gray-level matrix feature extraction algorithm was used to extract the features of pump dynamometer card,and gray eigenvalues were up-sampled by a synthetic minority oversampling technique(SMOTE)to achieve the equalization of unbalanced data.A dragonfly algorithm was used to select RF parameters and recognize the operating conditions of pumping wells,and the experimental verification was performed using the production data of Liaohe oilfield.The results demonstrate that the as-proposed method can avoid the inaccurate selection of reference operating conditions with traditional identification method,reduce the influence of unbalanced data on the identification of operating condition,improve the identification accuracy of operating condition and meet the actual demand of oilfield.

关 键 词：有杆抽油系统 示功图 随机森林 蜻蜓优化算法 过采样技术 工况识别 不平衡数据集 灰度特征值 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]