Refork Completion Analysis with the Aid of Artificial Neural Networks

Abstract

Abstract The quality and quantity of information available in the public domain is growing rapidly. Companies are creating in-house databases to track and improve operating and service performances in an effort to keep up with this explosion of data. The hardware and software used to obtain and manipulate massive amounts of information are constantly improving. All these events have created an opportunity to evaluate the complex interaction of variables and quantify how they relate to the required end result. The Redfork formation is a prolific, low-permeability, natural gas and gas-condensate reservoir deposited during the middle Pennsylvanian Period. The reservoir is located in the deep Anadarko basin of west-central Oklahoma. The Redfork is an interesting reservoir because of its high level of heterogeneity and the varied stimulation/completion methods used in the formation. The volume and diversity of information available on Redfork completions make the Redfork formation unique. This paper analyzes Redfork completions in Roger Mills and Custer Counties. The study uses artificial neural networks (ANNs) trained on a data set of 107 Redfork completions to analyze and quantify the effect of well/reservoir parameters and completion methods on production results. Specific areas of interest include controllable/quantifiable aspects of a well's completion/stimulation procedure that affect the production outcome. The paper will document a methodological test that supplements standard completion-optimization techniques. P. 555