ABSTRACT

Predicting well performance and production prior to drilling has always been a challenge in the oil and gas industry. The need of developing robust methods of prediction helps it to be financially more viable and technically superior. With the emergence of big data, the ever-increasing computational power, and the growing library of platforms supporting a multitude of machine learning methods, data and analytics departments can play a crucial role in the development of fact-based methodologies and predictive models of future performance with high levels of confidence. In this paper we lay a series of supervised regression methods that proved highly predictive when compared to existing methods of estimating undrilled well production. We conducted an extensive machine learning modeling exercise using data from an active Jonah Energy (Jonah) gas field in Sublette County, Wyoming. Our objective was to predict well performance and annual-measured gas production in the first year of production. We considered several multidimensional and complex data (geological, drilling and production data) as direct and indirect factors that control production. We used a fact-based methodology for feature selection that focused on the use of measurement data rather than estimations or interpretation. Finally, we tested and compared various machine learning algorithms-including Linear Regression, Principal Component Analysis, Neural Network Regression, Boosted Decision Tree, and Binned Decision Tree-to find the optimum prediction of the current gas field. Ultimately, we judged the best results resulted by using Binned Decision Tree where the Coefficient of Determination was 0.63 and squared error was 0.36. Further Ensemble Decision forest was also tested to rule out any overfitting. Given the complexity of geology and physical limitation present dataset, the current production prediction seems prudent.

Presentation Date: Tuesday, September 17, 2019

Session Start Time: 8:30 AM

Presentation Time: 10:10 AM

Location: 221C

Presentation Type: Oral

This content is only available via PDF.
You can access this article if you purchase or spend a download.