Applying a single scaling factor over the life of the well for a completions variable will give you an inaccurate production profile — and inaccurate economics!
Quantifying the time-variant impact of completions decisions is tough for traditional methods. However, training a machine learning model to predict a time series of production, combined with Shapley values, provides a powerful tool to investigate when completions are most impactful.
Wednesday, 9:30 AM. Theme 8: EUR and Performance Prediction: Decline Curve Analysis and Beyond III
SHAP values for proppant, fluid loading, stage length, geology, and spacing, for tightly-spaced wells with >1000 #/ft proppant in the Williston Basin. Proppant shows most impact from IP120-210, with steadily decreasing importance through time.
In unconventional plays, operators commonly use scaling factors to adjust actual or expected production between groups of wells with different completions designs—e.g., increasing proppant loading from 1500 lbs/ft to 1750 lbs/ft should see 6% production uplift. Though these scalars are easily deployed, they suffer inaccuracy in the time domain away from the date at which the analysis was anchored. Here, we present a machine learning-based study of the Bakken-Three Forks play of the Williston Basin, showing that large completions designs have the biggest impact on production between IP days 90-180, with the impact steadily decreasing through time afterwards. This method can be used to build scaling factors for any completions or spacing parameter by using SHAP values (SHapley Additive exPlanations), which isolate the contribution of each feature on the model prediction. Proppant loading, fluid loading, and stage length all show strong variation in scalar impact through time. All three parameters show diminishing impact over the life of the well, with large designs showing approximately 55% uplift in rates over average designs at IP90 but only 30% uplift over small designs by IP720. In contrast, the relative importance of inter-well spacing and geology increases steadily through time. SHAP values offer a powerful method to extract scaling factors from a tree-based machine learning model. Because they can be incorporated into the model-building pipeline, they remove the need to run synthetic cases or build partial dependence plots. Applying time-dependent scaling factors when making well predictions or building type curves will result in a more accurate production profile, improving decision making no matter whether the operator prefers payback or rate of return economic metrics. These methods can also be used to help answer to what extent intense designs or tight spacing accelerates or improves recovery.