基于可解释性USL的抽油井电功率异常检测
Anomaly Detection of Pumping Well Motor Data Based on Interpretable USL
袁春华 1冯博文 1李翔宇1
作者信息
- 1. 沈阳理工大学 自动化与电气工程学院,沈阳 110159
- 折叠
摘要
抽油井长期运行可能使有杆泵抽油系统出现异常工况,这种异常的产生与设备的持续退化有关,但其具体原因仍难以通过数据驱动的深度学习方法来分析和解释.为了能够及时准确地检测到异常工况并推断故障级别和故障类型,提出一种基于可解释性无监督Shapelet学习(unsupervised Shapelet learning,USL)算法的抽油井电功率异常检测方法.首先利用提取到的先验知识改进USL;然后将改进的USL与单分类器结合,构成基于可解释性USL的异常检测算法(AD-IUSL),AD-IUSL可同时学习Shapelet特征和异常阈值.利用具有不同异常工况的抽油井电功率数据进行验证,结果表明,本文提出算法的F1 分数达到了 0.969,根据学习到的Shapelet特征合理解释了两种异常发生的潜在原因.
Abstract
Long-term operation of oil wells can lead to abnormal conditions in the sucker rod pump system.These abnormalities are associated with the continuous degradation of equipment.However,it is still difficult to analyse and interpret the specific causes using data-driven deep learning meth-ods.To detect such abnormal conditions promptly and accurately and to deduce the level and type of faults,an anomaly detection method for the motor power of oil wells is proposed based on an in-terpretable unsupervised Shapelet learning(USL)algorithm.Initially,the USL is improved using extracted prior knowledge.Subsequently,the USL is combined with a one-class classifier to form an anomaly detection algorithm based on interpretable USL(AD-IUSL),which can simultaneously learn Shapelets and anomaly thresholds.This study validates the proposed method using electrical power data from oil wells with two different abnormal conditions.Experimental results demonstrate that the proposed algorithm achieves an F1-score of 0.969 and can explain the potential reasons for two types of anomalies based on the learned Shapelet feature.
关键词
抽油井/异常检测/可解释性/Shapelet学习算法Key words
pumping well/anomaly detection/interpretability/Shapelet learning algorithm引用本文复制引用
基金项目
国家自然科学基金项目(62173073)
辽宁省教育厅高等学校基本科研项目(LJKMZ20220618)
辽宁省教育厅高等学校基本科研项目(JYTMS20230215)
辽宁省本科教改优质教学资源建设与共享项目(SBKJGYZ-2021-06)
出版年
2024
NETL
NSTL
万方数据