Machine learning based approach to detect well analogue

1. A method comprising:

collecting a training log from a plurality of well logs;

creating a log window of a plurality of log windows from the training log;

creating a positive window from the log window;

creating a negative window from the training log; and

training a siamese neural network (SNN) that includes a first self attention neural network (ANN) and a duplicate self attention neural network with the log window, the positive window, and the negative window, to recognize a similarity between the log window and the positive window and to differentiate against the negative window, wherein training the siamese neural network comprises generating the first self attention neural network output of the first self attention neural network using a plurality of convolutional neural networks.

2. The method of claim 1 , wherein training the siamese neural network comprises:

generating a first self attention neural network output with the first self attention neural network in response to inputting the log window into the first self attention neural network;

generating a duplicate self attention neural network output with the duplicate self attention neural network in response to inputting a probe window into the duplicate self attention neural network, wherein the duplicate self attention neural network is within the siamese neural network and is a duplicate of the first self attention neural network;

generating a distance layer output in response to inputting the first self attention neural network output and the duplicate self attention neural network output to a distance layer of the siamese neural network; and

generating a siamese neural network output from the distance layer output using an output layer of the siamese neural network.

3. The method of claim 2 , wherein training the siamese neural network comprises:

comparing a siamese neural network output to a label of a probe window to form a siamese neural network output comparison; and

updating the siamese neural network based on the siamese neural network output comparison using backpropagation with the first self attention neural network and the duplicate self attention neural network receiving similar updates.

4. The method of claim 1 , further comprising:

selecting a query window of a plurality of query windows;

selecting a well log of a target well;

splitting the well log into a plurality of target windows;

comparing the query window to a target window of the plurality of target windows using the siamese neural network with the first self attention neural network after training to generate a result output; and

presenting a result that includes the target window and the result output.

5. The method of claim 1 , further comprising:

tagging a query window with a tag from one of a group comprising a target window, a well log of the target window, and a target well of the target window.

6. The method of claim 1 , further comprising:

tagging one of a group comprising a target window, a well log of the target window, and a target well of the target window with a tag from a query window.

7. A method comprising:

collecting a query window of a plurality of query windows;

collecting a well log of a target well;

splitting the well log into a plurality of target windows;

comparing the query window to a target window of the plurality of target windows using a siamese neural network (SNN) that includes a self attention neural network (ANN) to generate a result output, wherein the siamese neural network is trained with a log window, a positive window, and a negative window, to recognize a similarity between the log window and the positive window and to differentiate against the negative window; and

presenting a result that includes the target window and the result output.

8. The method of claim 7 , further comprising:

tagging the query window with a tag from one of a group comprising the target window, the well log of the target window, and the target well of the target window.

9. The method of claim 7 , further comprising:

tagging one of a group comprising the target window, the well log of the target window, and the target well of the target window with a tag from the query window.

10. The method of claim 7 , further comprising:

wherein training the siamese neural network comprises:

generating a first self attention neural network output with the self attention neural network in response to inputting a training window into the self attention neural network;

generating a duplicate self attention neural network output with a duplicate self attention neural network in response to inputting a probe window into the duplicate self attention neural network, wherein the duplicate self attention neural network is within the siamese neural network and is a duplicate of the self attention neural network;

generating a distance layer output in response to inputting the first self attention neural network output and the duplicate self attention neural network output to a distance layer of the siamese neural network; and

generating a siamese neural network output from the distance layer output using an output layer of the siamese neural network.

11. The method of claim 7 , further comprising:

wherein training the siamese neural network comprises:

comparing a siamese neural network output to a label of a probe window to form a siamese neural network output comparison; and

updating the siamese neural network based on the siamese neural network output comparison using backpropagation.

12. A system comprising:

a memory coupled to a processor; and

a training application that executes on the processor, uses the memory, and is configured for:

collecting a training log from a plurality of well logs;

creating a log window of a plurality of log windows from the training log;

creating a positive window from the log window;

creating a negative window from the training log; and

training a siamese neural network (SNN) that includes a first self attention neural network (ANN) and a duplicate self attention neural network with the log window, the positive window, and the negative window, to recognize a similarity between the log window and the positive window and to differentiate against the negative window, wherein training the siamese neural network comprises generating the first self attention neural network output of the first self attention neural network using a plurality of convolutional neural networks.

13. The system of claim 12 , wherein training the siamese neural network comprises:

generating a first self attention neural network output with the first self attention neural network in response to inputting the log window into the first self attention neural network;

generating a duplicate self attention neural network output with the duplicate self attention neural network in response to inputting a probe window into the duplicate self attention neural network, wherein the duplicate self attention neural network is within the siamese neural network and is a duplicate of the first self attention neural network;

generating a distance layer output in response to inputting the first self attention neural network output and the duplicate self attention neural network output to a distance layer of the siamese neural network; and

generating a siamese neural network output from the distance layer output using an output layer of the siamese neural network.

14. The system of claim 12 , wherein training the siamese neural network comprises:

comparing a siamese neural network output to a label of a probe window to form a siamese neural network output comparison; and

updating the siamese neural network based on the siamese neural network output comparison using backpropagation with the first self attention neural network and the duplicate self attention neural network receiving similar updates.

15. The system of claim 12 , further comprising:

a server application that executes on the processor, uses the memory, and is configured for:

selecting a query window of a plurality of query windows;

selecting a well log of a target well;

splitting the well log into a plurality of target windows;

comparing the query window to a target window of the plurality of target windows using the siamese neural network with first the self attention neural network after training to generate a result output; and

presenting a result that includes the target window and the result output.

16. The system of claim 12 , further comprising:

a server application that executes on the processor, uses the memory, and is configured for:

tagging a query window with a tag from one of a group comprising a target window, a well log of the target window, and a target well of the target window.

17. The system of claim 12 , further comprising:

a server application that executes on the processor, uses the memory, and is configured for:

tagging one of a group comprising a target window, a well log of the target window, and a target well of the target window with a tag from a query window.