Causal Analysis and Data Mining of Well Stimulation Data Using Classification and Regression Tree with Enhancements

Abstract

The well treatment program is an important part of the field development plan, and certain variables, such as job pause time (JPT) and fracture screenout, can affect its efficiency. JPT is the time during which pumping is paused in between subsequent treatments of a job. Screenout occurs because of a sudden restriction of fluid flow inside the fracture and through the perforation. The objectives of this work are to investigate whether, from existing data, it is possible to find patterns in significant variables that affect the extreme values of JPT in a particular region and what is the most critical variable causing screenout. The answers are sought by applying a classification and regression tree (CART) to both categorical and continuous variables in the database. The practical application of CART is presented using case studies containing JPT and screenout. Significant variables are found that affect the response variables, and predictor variables are ranked in the hierarchal order of their importance. Such information can be used to control predictor variables that cause high JPT or screenout. The results are outlined in an intuitive way, including categorical, continuous, and missing values. Because CART is a data driven, deterministic model, one cannot calculate the confidence interval of the predicted response. The confidence in results is purely based on the historical values, and the accuracy of the result produced by a tree model depends on the quality of recorded data measured in terms of volume, reliability, and consistency. The prediction capability of CART is enhanced by using the normal score transform (NST) and by dividing the large dataset into smaller groups using clustering. The approach presented in this paper analyzes a dataset under limited information and high uncertainty and should lead to developing methodology for generating proxy models to find future success indices (e.g., one for drilling efficiency or production from a fracture). This could standardize stimulation and generate decision practices to save costs in field development and the optimization process.

Multivariate Analysis and Data Mining of Well-Stimulation Data by Use of Classification-and-Regression Tree With Enhanced Interpretation and Prediction Capabilities

“…In this paper, the CART methodology is first introduced, followed by the application on four case studies by use of a data compilation that includes general well and job information, joblevel summary data, pumping-schedule stage-level summary data, pumping-schedule individual-stage data including additives, wellbore and completion data, event-logger data, and equipment data, collected from more than 200,000 fracturing and data-acquisition jobs from all over North America since 2004 (Bhattacharya et al , 2014. Further, an approach is presented to enhance the predictive capability of the CART method by additional data preprocessing by use of normal-score transform (Everitt 2002) and kernel-means clustering (Schölkopf et al 1996;Shawe-Taylor and Cristianini 2004).…”