Training data for machine learning seismic inversion

Patent metadata

Field	Value
Publication number	US-12130398-B2
Application number	US-202217672486-A
Country	US
Kind code	B2
Filing date	Feb 15, 2022
Priority date	Feb 15, 2022
Publication date	Oct 29, 2024
Grant date	Oct 29, 2024

Abstract

Well data (e.g., well log) may be divided into multiple segments, and different samplings of data in the individual segments may be performed to increase the amount of data that is used to train a seismic inversion model. Synthetic well data may be generated from real well data to increase the amount of well data from which sampling is performed.

First claim

What is claimed is:

1. A system for improving seismic inversion using improved training data, the system comprising:
one or more physical processors configured by machine-readable instructions to:
obtain well information, the well information defining subsurface configuration of wells, the subsurface configuration of wells defined by values of multiple subsurface properties;
divide individual subsurface configuration of wells defined by the well information into multiple segments, individual segments including a portion of the subsurface configuration of wells defined by the well information;
generate the improved training data for a seismic inversion model based on sampling of the individual segments, wherein the sampling of the individual segments includes separate random sampling with replacement of the multiple subsurface properties, wherein the separate random sampling with replacement of the multiple subsurface properties is performed for a top layer and a bottom layer;
train the seismic inversion model using the improved training data; and
store the trained seismic inversion model in a non-transient storage medium.

2. The system of claim 1 , wherein the one or more physical processors are further configured by the machine-readable instructions to:
obtain seismic reflection information, the seismic reflection information defining seismic reflection response in a subsurface region; and
determine subsurface characteristics in the subsurface region by performing seismic inversion of the seismic reflection response in the subsurface region using the trained seismic inversion model.

3. The system of claim 2 , wherein the seismic reflection response is calibrated and seismic attributes are computed from the calibrated seismic reflection response, further wherein the trained seismic inversion model is applied to the seismic attributes to perform the seismic inversion of the seismic reflection response in the subsurface region.

4. The system of claim 3 , wherein the improved training data includes pairings of corresponding seismic attributes and subsurface characteristics.

5. The system of claim 4 , wherein the subsurface characteristics include acoustic impedance, shear impedance, and density.

6. The system of claim 1 , wherein the individual segments cover a range of tens to hundreds of feet.

7. The system of claim 6 , wherein the sampling of the individual segments to generate the improved training data for the seismic inversion model is performed hundreds to thousands of times for the individual segments.

8. The system of claim 1 , wherein the wells include one or more synthetic wells, and the subsurface configuration of the wells defined by the well information includes synthetic subsurface configuration of the one or more synthetic wells.

9. The system of claim 8 , wherein the one or more synthetic wells are generated from one or more real wells by varying sand properties of the one or more real wells while retaining shale properties of the one or more real wells.

10. A method for improving seismic inversion using improved training data, the method comprising:
obtaining well information, the well information defining subsurface configuration of wells, the subsurface configuration of wells defined by values of multiple subsurface properties;
dividing individual subsurface configuration of wells defined by the well information into multiple segments, individual segments including a portion of the subsurface configuration of wells defined by the well information;
generating the improved training data for a seismic inversion model based on sampling of the individual segments, wherein the sampling of the individual segments includes separate random sampling with replacement of the multiple subsurface properties, wherein the separate random sampling with replacement of the multiple subsurface properties is performed for a top layer and a bottom layer;
training the seismic inversion model using the improved training data; and storing the trained seismic inversion model in a non-transient storage medium.

11. The method of claim 10 , further comprising:
obtaining seismic reflection information, the seismic reflection information defining seismic reflection response in a subsurface region; and
determining subsurface characteristics in the subsurface region by performing seismic inversion of the seismic reflection response in the subsurface region using the trained seismic inversion model.

12. The method of claim 11 , wherein the seismic reflection response is calibrated and seismic attributes are computed from the calibrated seismic reflection response, further wherein the trained seismic inversion model is applied to the seismic attributes to perform the seismic inversion of the seismic reflection response in the subsurface region.

13. The method of claim 12 , wherein the improved training data includes pairings of corresponding seismic attributes and subsurface characteristics.

14. The method of claim 13 , wherein the subsurface characteristics include acoustic impedance, shear impedance, and density.

15. The method of claim 10 , wherein the individual segments cover a range of tens to hundreds of feet.

16. The method of claim 15 , wherein the sampling of the individual segments to generate the improved training data for the seismic inversion model is performed hundreds to thousands of times for the individual segments.

17. The method of claim 10 , wherein the wells include one or more synthetic wells, and the subsurface configuration of the wells defined by the well information includes synthetic subsurface configuration of the one or more synthetic wells.

18. The method of claim 17 , wherein the one or more synthetic wells are generated from one or more real wells by varying sand properties of the one or more real wells while retaining shale properties of the one or more real wells.

Assignees

• Chevron Usa Inc

Inventors

• Roy Prasenjit
• Zhang Yang
• Zhu Xinfa
• Cardona Reynaldo
• Hua Yong
• Fu Kun
• Almoughraby Bassel Mouloud

Classifications

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G01V1/003

  Seismic data acquisition in general, e.g. survey design (G01V1/3808, G01V1/42 take precedence) · CPC title

• G01V2210/1299

  Subsurface, e.g. in borehole or below weathering layer or mud line · CPC title

• G01V2210/62

  Physical property of subsurface · CPC title

• G01V1/46 Primary

  Data acquisition · CPC title