Application of Data Mining for Quick Root-Cause Identification and Automated Production Diagnostic of Gas Wells With Plunger Lift

Abstract

Summary The paper presents an application of the classification-and-regression-tree (CART) technique as a root-cause identification and production diagnostic tool, and presents case studies from real field data for gas wells using the plunger-lift system. Specifically, regression-tree analysis was performed on the basis of the available operation data. The regression-tree model is data-driven and easy to construct, and does not require all the parameters that a first-principles-based model often requires. Thus, these models are quite useful in cases where real-time data or the required data for a detailed analysis are unavailable. To improve prediction capability, a cross-validation technique was used to optimize tree size. A total of 8 to 10 variables that could potentially affect the gas-production rate was considered, with case studies conducted on actual field data. Questions, such as which group of wells is performing well or poorly, are quickly answered. Regression-tree analysis, when applied at the field level, can identify a group of wells that underperform compared with other wells. As a result, an asset team can prioritize wells identified as poor performers. Further, a statistical analysis helps to gain insight into understanding the causal relationship between the gas-production rate and various operational variables. On the basis of this analysis, recommendations are also provided to improve the gas-production rate of poorly performing wells. Individual-well models were also constructed to identify the root causes for high or low gas production on the basis of operational changes made during a certain period of time. Only a few of the variables were found to have a significant impact on gas production. The interpretation of the decision tree indicated that additional operational variables, such as production time and pressure-buildup time, should be included in the regression-tree analysis to improve the diagnostic process. This paper summarizes these conclusions and recommends future work to improve the prediction capability of the regression-tree analysis.

Prediction of vitrinite reflectance values using machine learning techniques: a new approach

“…This process is binary recursive partitioning that divides parent nodes into child nodes (binary splitting). This process is continued to reach terminal nodes that do not have any splitting (Singh 2017). Figure 23 shows the best tree for predicting vitrinite reflectance after running the decision tree model.…”

A Comparison of Machine Learning Approaches for Prediction of Permeability using Well Log Data in the Hydrocarbon Reservoirs

“…The name of this process is binary recursive partitioning. After that, parent nodes are divided into two child nodes whose name is binary splitting, and this process will be continued until arriving at nodes that will not have any splitting named terminal nodes [47]. Finding the splitting criteria, which depends on inputs, a DT begins, and then the value of square error between observed and calculated outputs is minimized.…”