Well log correlation system
A method can include receiving well logs, as series data, for wells, where the well logs represent stratigraphic characteristics of a field; for each of the well logs, generating a corresponding vector representation in a dimensional space using a transformer encoder; and determining similarity of the well logs via their corresponding vector representations to characterize well log quality for well log correlation of the field.
1. A method comprising:
drilling, using one or more drilling tools, a plurality of wells in a field;
generating, via one or more downhole sensors, a series of measurements indicative of stratigraphic characteristics of the field while drilling the plurality of wells;
receiving, as well logs, the series of measurements indicative of the stratigraphic characteristics of the field;
for each of the well logs, generating a corresponding vector representation in a dimensional space using a transformer encoder;
determining similarity of the well logs via their corresponding vector representations to characterize well log quality for well log correlation of the field, comprising implementing multi-dimensional scaling for the vector representations in the dimensional space to generate output in a reduced dimensional space;
rendering a visualization to a display of the output in the reduced dimensional space, the visualization comprises a similarity metric as an additional display attribute;
determining, based on the similarity of the well logs, that a first well included in the plurality of wells is an outlier and a second well included in the plurality of wells is an inlier; and
adjusting a first stratigraphic marker included in a first well log corresponding to the first well based on a second stratigraphic marker included in a second well log corresponding to the second well.
2. The method of claim 1 , wherein the transformer encoder comprises a bi-directional transformer encoder.
3. The method of claim 1 , wherein determining similarity of the well logs comprises implementing a Siamese neural network.
4. The method of claim 1 , wherein the similarity metric is color coded.
5. The method of claim 1 , wherein each of the well logs comprises a corresponding common stratigraphic marker for a common stratigraphic feature.
6. The method of claim 5 , wherein the common stratigraphic marker is a human identified stratigraphic marker.
7. The method of claim 5 , wherein the common stratigraphic marker is a machine identified stratigraphic marker.
8. The method of claim 7 , wherein determining similarity of the well logs via their corresponding vector representations to characterize well log quality is based on the common stratigraphic marker.
9. The method of claim 5 , further comprising determining a window size that encompasses the common stratigraphic marker.
10. The method of claim 9 , wherein the window size is defined with respect to time or with respect to depth.
11. The method of claim 1 , wherein determining similarity generates a normalized probability score of well log marker window similarity through implementation of a Siamese neural network.
12. The method of claim 1 , further comprising training the transformer encoder via self-supervised learning.