Release the APE: A Novel Approach to Produce Estimated Well Logs on Horizontal Wells and Improve Multivariate Predictive Models
Release the APE: A Novel Approach to Produce Estimated Well Logs on
Horizontal Wells and Improve Multivariate Predictive Models
Patrick Rutty & Chris W. Grant
Horizontal wells typically lack data because of the cost to run a full suite of well logs, yet they are surrounded by “forests” of vertical wells with full log suites. Using these vertical logs to directly estimate values at the horizontal wells makes it possible to improve the multivariate predictive models used to optimize engineering attributes in horizontals, such as lateral length, azimuth, proppant, and fluid.
To accomplish this, we use a direct estimation approach to extrapolate and interpolate well log data from vertical wells to horizontals. A structural model is used to guide the stratigraphic interpolation while honoring stratigraphic layering rules. This direct estimation produces a suite of logs at any specified resolution along the horizontals, or for any other type of wellbore. This approach benefits from accurately estimating the stratigraphic changes in a formation, as opposed to simplistic 2D methods that average each vertical well’s logs over an interval, then map these values, and then extract the relevant property from the resulting 2D grid.
This direct estimation approach also avoids the over-averaging typically used in 3D geocellular reservoir modeling. In that approach, 3D property modeling is performed on large 3D grids constructed from a surface-based geologic model. The 3D properties are extracted onto the wellbores to obtain a well log. The drawbacks of such a workflow are that they are time consuming, require specialized staff, over-average data, suffer from the discretization parameters, and can require large amounts of RAM and disk space. Our direct estimation approach suffers none of these problems because all these steps have been abstracted, obviating the need for a 3D grid. The result is a finely sampled log that honors both the surrounding vertical well logs and the stratigraphic layering scheme from the input geologic model, as well any seismic attribute inputs or other “soft” information (e.g., measurements while drilling) along the horizontal wellbores. Our approach is extremely fast and uses minimal RAM or disk space.
We use well log values from this direct estimation method in a multivariate case study in the Midland Basin in which we investigate production as a function of engineering and well log attributes. The improvements gained from using this approach (compared to the simplistic 2D method) include better correlation coefficients and increased significance of geologic variables in the model.
AAPG Datapages/Search and Discovery Article #90350 © 2019 AAPG Annual Convention and Exhibition, San Antonio, Texas, May 19-22, 2019