[URTeC 2022] How does the impact of completions change over the life of a well? A comparison across the major US unconventional plays using machine learning (ID 3723930)

How does the impact of completions change over the life of a well? A comparison across the major US unconventional plays using machine learning

Talk Details::
- Tuesday, July 21st at 2:40 PM | Room 361
- Theme 5: Geomechanics in Field Development (Well Placement, Drawdown, and Optimization)
- AUTHORS:: T. Cross (Novi Labs), K. Long, D. Niederhut, A. Cui

Abstract::

Objectives/Scope
Upsized completions have been responsible for a large increase in unconventional well productivity over the past decade. It is generally understood that the impact of upsized completions will be largest earlier in the life of a well, but quantifying the decay through time can be difficult with traditional methods. Multi-target machine learning models that forecast a time series of production, combined with explanation datasets like SHAP values, can provide a view into how different completions parameters impact well performance through time. In this study, we compare the time-variant impact of completions design across the Bakken, Eagle Ford, Niobrara (DJ), Delaware, Midland, Haynesville, Marcellus, and Barnett plays.

Methods/Procedures/Process
We compiled basin-wide datasets consisting of publicly available completions design parameters, production, interwell spacing, and geologic variables. We trained decision tree-based models to predict a vector of monthly cumulative production values for the first 3 years of the well life. We then generated SHAP Values (SHapley Additive exPlanations) that explain how much each training feature impacted the prediction (in bbl or mcf) at each timestep. To study basin-wide trends, we examined average SHAP values for proppant and fluid intensity, as stage spacing was not widely available. To facilitate basin-to-basin comparisons, we then calculated a “decay factor” that represents how much the impact decays from early time to 3 years, normalized to average well production rate decline in the basin.

Results/Observations/Conclusions
Over time, in all the plays analyzed, proppant and fluid have a greater reduction in impact on liquid production as compared to gas production. In addition, across most plays, proppant intensity’s impact on production decayed faster than that of fluid intensity. In the oil plays, the Delaware had the lowest decay, with 17% for proppant and 12% for fluid. By contrast, the Bakken had 54% decay for proppant and 37% for fluid. The gas plays and some gas production streams of oil plays showed increasing impact through time of completions design, with the Haynesville having 76% increasing impact of proppant and 114% increase in impact of fluid over the first three years.

Applications/Significance/Novelty
Engineers must take into account the time-varying impact of completions on their production forecasts and design decisions. Furthermore, this type of analysis can be used to analyze the interaction between the reservoir and the completions design. For instance, the faster decay of proppant relative to fluid across nearly all plays suggests that, over time, the impact of increased hydraulic fracture permeability (caused by proppant) decreases relative to fracture complexity and length (caused by fluid), with the exception of the Eagle Ford, which has high fracture smoothness and low matrix permeability. This workflow also generalizes to other design parameters, such as interwell spacing or artificial lift type.

Complete the form below and we will email the paper straight to your inbox.

First name(Required)
This field is for validation purposes and should be left unchanged.

INTRODUCING CAUSAL MODELS

Accurate forecast on parent-child developments

In this live webinar, you will learn how Novi’s new algorithm improves model sensitivity for spacing and parent-child scenarios, providing powerful results for previously difficult-to-analyze problems.

Ted Cross, our VP of Product Management, will show you how this update improves spacing and infill scenario analysis without sacrificing model accuracy.