--> Abstract: Data Mining Well Completion Data for the Dakota Formation, San Juan Basin, New Mexico, by R. Balch and A. K. Iduri; #90092 (2009)

Datapages, Inc.Print this page

Data Mining Well Completion Data for the Dakota Formation, San Juan Basin, New Mexico

Robert Balch1 and Ajay K. Iduri2
1Petroleum Recovery Research Center, New Mexico Tech, Socorro, NM
2Computer Science Department, New Mexico Tech, Socorro, NM

Data mining was used to analyze completion data for a tight gas data set to identify trends or interesting patterns between well completion/stimulation methods and gas production. The study data set was 370 non-commingled Dakota wells completed between 1994-2004. Predictive models were tested using 58 Dakota wells completed between 2004-2006. The project used data donated by IHS Energy. Data included geographical attributes Company Name, Completion Date, Location, and Depth to Dakota Top and non-geographical attributes Fracture Stages, Fracture Net Thickness, Fracture Gross Thickness, Fracture Fluid Type, Sand Lbs, Sand Type, Sand Size, and Sand Additive.

Differences between well successes by company were evaluated first. First year’s gas (FYG) was selected as production indicator to compare wells with varying production time. A two-sample T-Test was performed with a null hypothesis that each company would match the average of all companies. Six of eight companies were statistically different from the null hypothesis. Attempts to cluster using location and production data by company resulted in no dominant trends. Completion/stimulation data was then mined to find the best parameters for predicting FYG. Hypothesis-generating approaches discovered interesting relationships and patterns in the data and each technique identified Fractured Fluid Gallons, Fractured Gross Thickness, Fractured Fluid Type, Sand Lbs, and Acid Gallons as key attributes.

Predictive models were built using regression trees and neural networks to predict FYG using these attributes. The best model used Fracture Net Thickness, Fracture Fluid Gallons, and Sand lbs and formed a nonlinear regression with 87 coefficients and correlation coefficients of 0.93 and 0.84 for training and testing data, respectively. The model accurately predicted FYG at 58 wells not included in the analysis.

AAPG Search and Discovery Article #90092©2009 AAPG Rocky Mountain Section, July 9-11, 2008, Denver, Colorado