[URTeC 2020] The Impact of Spacing and Time on Gas/Oil Ratio in the Permian Basin: A Multi-Target Machine Learning Approach (ID 2676)

As unconventional, horizontal development in the Permian Basin moves beyond adolescence into middle age, operators will increasingly have to deal with older and older parent wells. What does that mean for GOR?

 

The older the parent, the higher the GOR of the child, in the eyes of our model. We tackle that topic, along with distance to parent, and more, in this paper.

TALK DETAILS:

Monday, 1:45 PM. Theme 5: Integrated Geochemistry of Oil-Prone and Gas-Prone Unconventional Resource Plays with Geology and Petrophysics II

Percent Impact on Gas and Oil Production, as measured with Shapley values, of the number of days the parent has been online. Dashed line is impact at 90 production days, solid is at 720 days, both cumulative.

ABSTRACT::

The gas/oil ratio (GOR) of unconventional wells in the Midland Basin is becoming an increasingly important topic, due to both transportation capacity and the general price conditions for gas products. Coupled with this, interwell spacing tests have caused many to rethink the economic viability of certain parts of shale basins. This study presents a machine learning approach to predicting GOR in the Midland Basin over the first 2 years of a well’s production history. 

We utilize a dataset compiled from public sources to make predictions of oil, gas, and water production for all horizontal unconventional wells in the Midland basin. The prediction outputs are a time series at 30 day increments, so that we can observe the decline rates of each fluids product with time. This multi-input-multi-output (MIMO) model utilizes completions information, interwell spacing, parent child calculations, and geology information sourced from well logs. 

Using the results of the model, we generate Shapley values, a method of interrogation which model features (interwell spacing, proppant loading, porosity, etc) had the greatest impact on the predicted production profiles. This method isolates the impacts of geological factors from completions intensity, and in turn, interwell spacing. 

Out of the dozens of spacing calculations utilized in this model, the number of codeveloped sibling wells carried the strongest signal for the total oil production, and the number of parents in the surrounding area was the strongest driver of gas production. Unsurprisingly, wells with 5+ codeveloped neighbors negatively impacted by as much as 10% relative to unbounded wells. Gas production showed the opposite relationship, showing 15-20% increases moving for wells with completed in areas with 5+pre-existing parents. Taken together, these effects compound to cause significant increases in GOR. Furthermore, the increased GOR is not apparent in the early days of the predicted decline curves. For wells with 5+ neighbors, we predict GORs to up to 10 times higher at IP720 than IP30. 

SHAP values from a multi-target machine learning model can be used to predict the impact of interwell spacing on oil and gas production. Crucially, this approach shows that damaging effects of interwell spacing on liquids production may not reveal itself until several months into a well’s production.  In fact, these model SHAP values show that in many cases, there may be no damage to production rates in the first 3 months of a downspaced child wells, but that oil production will decrease in conjunction with increased gas production significantly after that.  These insights are crucial to maintaining low GOR and maximizing asset value of child wells. 

Enter your name and email below. We will email the paper to you right away.

First name(Required)

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.