	US 12,072,459 B2
	Subsurface property estimation in a seismic survey area with sparse well logs
Haibin Di, Houston, TX (US); Xiaoli Chen, Houston, TX (US); Hiren Maniar, Houston, TX (US); and Aria Abubakar, Sugar Land, TX (US)
Assigned to SCHLUMBERGER TECHNOLOGY CORPORATION, Sugar Land, TX (US)
Appl. No. 17/758,514
Filed by SCHLUMBERGER TECHNOLOGY CORPORATION, Sugar Land, TX (US)
PCT Filed Jan. 11, 2021, PCT No. PCT/US2021/012886 § 371(c)(1), (2) Date Jul. 8, 2022, PCT Pub. No. WO2021/142406, PCT Pub. Date Jul. 15, 2021.
Claims priority of provisional application 62/959,374, filed on Jan. 10, 2020.
Prior Publication US 2023/0026857 A1, Jan. 26, 2023
Int. Cl. G01V 1/28 (2006.01); G01V 1/30 (2006.01); G01V 1/34 (2006.01); G01V 1/48 (2006.01); G06N 3/045 (2023.01); G06N 3/08 (2023.01)

CPC G01V 1/282 (2013.01) [G01V 1/306 (2013.01); G01V 1/345 (2013.01); G01V 1/48 (2013.01); G06N 3/045 (2023.01); G06N 3/08 (2013.01); G01V 2210/74 (2013.01)]

19 Claims

1. A method, comprising:

receiving, at a first machine learning model comprising an 85-layer convolutional auto-encoder, seismic data representing a subsurface domain in a given area as first training data for the first machine learning model;

training the first machine learning model utilizing the seismic data as training data;

extracting, using the first machine learning model, one or more first seismic features from the seismic data received at the first machine learning model, the seismic data representing the subsurface domain in the given area;

transmitting the one or more first seismic features extracted by the first trained machine learning model as a first portion of second training data;

receiving, at a second machine learning model comprising a deep neural network, the one or more first seismic features extracted by the first trained machine learning model as the first portion of the second training data;

receiving, at the second machine learning model, one or more well logs from one or more existing wells located at respective locations in the given area, the one or more well logs representing one or more subsurface properties in the subsurface domain as a second portion of the second training data;

receiving, at the second machine learning model, the seismic data representing the subsurface domain in the given area as a third portion of the second training data;

training the second machine learning model utilizing the first portion of the second training data, the second portion of the second training data, and the third portion of the second training data;

receiving, at the first machine learning model, second seismic data;

extracting, using the first machine learning model, one or more second seismic features from the second seismic data received at the first machine learning model;

transmitting the one or more second seismic features to the second machine learning model;

predicting, using the second machine learning model, the one or more subsurface properties in the subsurface domain that would normally be captured in a well log at a second location in the given area at which no well log data is available and that does not correspond to any of the respective locations of the one or more existing wells in the given area by making connections between the second seismic data received at the second machine learning model, the one or more well logs, and the one or more second seismic features that were extracted from the second seismic data by the first machine learning model and transmitted to the second machine learning model; and

performing at least one drilling operation of a well to be undertaken at the second location in the given area that does not correspond to any of the respective locations of the one or more the existing wells based on the one or more subsurface properties predicted by the second machine learning model.