document preview

Abstract
This paper documents a data-mining study of well, hydraulic fracture treatment, and production parameters for horizontal wells in the north Texas Barnett Shale play. In this study, the authors have analyzed well and production data from more than 13,400 producing Barnett wells. A subsample of over 3,300 horizontal wells was characterized with respect to detailed well architecture data such as drift direction and angle, lateral length, perforations, etc. The study uses Geographical Information System pattern-recognition techniques in conjunction with more traditional statistical techniques to interpret hidden trends in otherwise scattered data sets. This work provides a case study in the practical use of data-mining techniques to address questions of best practices in Shale Gas reservoirs. It is made possible because the availability and quality of public domain well and production data has increased significantly in the past few years. Simple cross plotting of production data against well and treatment variables normally leads to broad scattering of results. This study takes advantage of the largest, richest well and production data set available from the gas shales and identifies key lessons-learned. Relevant trends, such as the impact of toe up versus flat versus toe down, horizontal well length, and drift angle variability on gas production rate are presented. This work is significant in that it shows that the application of practical data-mining methods to a large Shale Gas data set can result in learning key lessons that may not be apparent when working with small data sets. This work is also significant in the use of merged reservoir quality proxies, well architecture data, completion data, and stimulation data, against which production results are placed in geographical perspective for improved interpretation.

Introduction
Shale reservoir drilling, completion, and stimulation have been the subject of much discussion in the industry over the past several years and discussion among industry workers as to what constitutes “best practice” is often spirited. This paper describes an ongoing effort to use data mining techniques to deconvolve effects of reservoir quality, well architecture, completion, and stimulation on well productivity in the Barnett Shale of the Fort Worth Basin, North Texas. Data sets from available Barnett Shale wells across the basin were prepared, examined for outliers and other data-quality issues, and analyzed using geographical information systems (GIS) and statistical techniques. Pattern recognition is recognized by the authors as an important tool in the analysis of large, multivariate data sets such as this one, where non-sympathetic, multi-variable impacts on production outcomes generate complex relationships that may be well-hidden in the data.

The paper provides an overview of Barnett Shale gas production and then reports on the analysis of well architecture effects on production results. Directional surveys from over 3,300 wells are used to characterize lateral length, azimuth, drift angle, and degree of undulation (porpoising), and to analyze and interpret impact on well productivity. The study shows that there are “global” and “local” lessons to be learned across the Barnett. In particular, it shows that Barnett horizontals become statistically less efficient as lengths increase beyond about 3,500 – 4,500 feet. The study also documents that the optimum Barnett drilling azimuth in the play is approximately 140 or 320 degrees, but cautions the reader that it is important to study and understand local stress changes and not to blindly drill on a statistical azimuth. Surprisingly, the study shows that well undulation or porpoising does not impact productivity in early well life.