S. Sharaf, P. Bangert, Mohamed Fardan, Khalil Alqassab, M. Abubakr, Mahmood Ahmed
{"title":"Beam Pump Dynamometer Card Classification Using Machine Learning","authors":"S. Sharaf, P. Bangert, Mohamed Fardan, Khalil Alqassab, M. Abubakr, Mahmood Ahmed","doi":"10.2118/194949-MS","DOIUrl":null,"url":null,"abstract":"A beam or a sucker rod pump is an artificial-lift pumping system using a surface power source to drive a downhole pump assembly. A beam and crank assembly creates reciprocating motion in a sucker-rod string that connects to the downhole pump assembly. The pump contains a plunger and valve assembly to convert the reciprocating motion to vertical fluid movement. A dynamometer is a diagnostic device used on sucker rod pumped wells that measures the load on the top rod and plots this load in relation to the polished rod position as the pumping unit moves through each stroke cycle.\n The analysis of the dynamometer card data delivers valuable insights on the status of the pump and indicates if future actions are required. In practice, the load versus displacement plot shape can be visually categorized in different classes where each shape has a specific meaning and indicates certain operating conditions. Machine learning algorithms are computing systems that learn to perform tasks by considering examples, generally without being programmed with any task-specific rules. During a period of approximately two (2) months, we collected 5,380,163 different cards from 297 beam pumps deployed in the Bahrain Field using the Supervisory Control and Data Acquisition (SCADA) system with an Open Platform Communication (OPC) interface. 35,292 cards are manually labelled by experts into twelve (12) classes. The dataset is split into 80% training and 20% holdout datasets. A training dataset is split into 5-fold cross validation. Different machine learning algorithms are evaluated predicting pump card class and their performance is compared. The top performing model, Gradient Boosting Machines (GBM) Classifier, achieves 99.98% accuracy in cross validation and 100% accuracy on holdout dataset without any extensive feature engineering.\n This paper explains the steps taken to improve surveillance of beam pumps using dynamometer card data and machine learning techniques and the lessons learned from executing the first Artificial Intelligence (AI) project within Tatweer Petroleum.","PeriodicalId":11321,"journal":{"name":"Day 3 Wed, March 20, 2019","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2019-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"5","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Day 3 Wed, March 20, 2019","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.2118/194949-MS","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 5
Abstract
A beam or a sucker rod pump is an artificial-lift pumping system using a surface power source to drive a downhole pump assembly. A beam and crank assembly creates reciprocating motion in a sucker-rod string that connects to the downhole pump assembly. The pump contains a plunger and valve assembly to convert the reciprocating motion to vertical fluid movement. A dynamometer is a diagnostic device used on sucker rod pumped wells that measures the load on the top rod and plots this load in relation to the polished rod position as the pumping unit moves through each stroke cycle.
The analysis of the dynamometer card data delivers valuable insights on the status of the pump and indicates if future actions are required. In practice, the load versus displacement plot shape can be visually categorized in different classes where each shape has a specific meaning and indicates certain operating conditions. Machine learning algorithms are computing systems that learn to perform tasks by considering examples, generally without being programmed with any task-specific rules. During a period of approximately two (2) months, we collected 5,380,163 different cards from 297 beam pumps deployed in the Bahrain Field using the Supervisory Control and Data Acquisition (SCADA) system with an Open Platform Communication (OPC) interface. 35,292 cards are manually labelled by experts into twelve (12) classes. The dataset is split into 80% training and 20% holdout datasets. A training dataset is split into 5-fold cross validation. Different machine learning algorithms are evaluated predicting pump card class and their performance is compared. The top performing model, Gradient Boosting Machines (GBM) Classifier, achieves 99.98% accuracy in cross validation and 100% accuracy on holdout dataset without any extensive feature engineering.
This paper explains the steps taken to improve surveillance of beam pumps using dynamometer card data and machine learning techniques and the lessons learned from executing the first Artificial Intelligence (AI) project within Tatweer Petroleum.