Abstract
Existing wellbores represent a resource that must be properly managed for a maximum return on investment. Many of the initial completions on existing wellbores are, in some manner, deficient: stimulation is inadequate, wellbore communication with the reservoir has not been achieved, productive pay has not been included in the initial completion, or well mechanics are inadequate. Identifying restimulation candidates can be a time-consuming process because of limited personnel and resources, variability in reservoir quality, incomplete data in well files, and the lack of data for conventional wellassessment procedures. In this paper, we discuss how an artificial neural network (ANN) analysis of public information can help identify well restimulation candidates with the highest potential for production improvement.
The subject of this paper is an ANN analysis of Red Oak restimulations in Oklahoma's Latimer and Le Flore counties. Readily available and measurable well attributes were selected as inputs for ANN training. These inputs included geographical location, surface elevation, initial and current reservoir pressure estimates, perforation location, initial completion procedures, refracture procedures, current production, etc. Relevant conclusions and generalizations quantify the effect of these reservoir, well, and completion factors on the potential for production improvement. This paper documents the development of an ANN that predicted Red Oak postrecompletion production with an average absolute error less than 5% (an R2 of 0.99).
