[go: up one dir, main page]
Scribd
Upload
51 views26 pages

Machine Learning in Oil & Gas Applications

The document discusses the application of machine learning in the oil and gas industry, focusing on shale image analysis, feature extraction, and the prediction of in-situ elastic properties. It outlines a workflow that includes data cleaning, feature engineering, and supervised learning techniques for predicting water production and generating geomodels. Various machine learning methods and feature distributions are analyzed to enhance the understanding of reservoir characteristics and improve production outcomes.

Uploaded by

MSlini
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
51 views26 pages

Machine Learning in Oil & Gas Applications

The document discusses the application of machine learning in the oil and gas industry, focusing on shale image analysis, feature extraction, and the prediction of in-situ elastic properties. It outlines a workflow that includes data cleaning, feature engineering, and supervised learning techniques for predicting water production and generating geomodels. Various machine learning methods and feature distributions are analyzed to enhance the understanding of reservoir characteristics and improve production outcomes.

Uploaded by

MSlini
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Machine Learning Applications in the Oil and Gas Industry

Shale Image Analysis and Connectivity


Quantification
Training and Testing Pixels

Wolfcamp map: 1000 image slices (2058 X 2606 pixels)


Barnett map: 56 image slices (2058 X 2606 pixels)

3
Feature Extraction
Features extracted
… ▪ Wavelet Transform (6)
▪ Gaussian Blur (1)
▪ Difference of Gaussian (1)

16 Features ▪ Hessian Matrix (3)


▪ Sobel Edge Detector (1)
▪ Local information (3)
Feature Vector
One Pixel … ▪ Pixel intensity (1)

4
Training and Testing Dataset
16-Dimensional Features Targets (Multiclass vs. Multilabel)

Pyrite (label 4)

Kerogen (label 2)
30k Samples

Pores/cracks (label 1)

Pyrite (label 4)

Rock matrix (label 3)

Kerogen (label 2)

Pores/cracks (label 1)

5
Connectivity of Constituents

6
Connectivity of Kerogen

Average connected distances of 125.97 and 25.48 Average travel times of 24.61 and 6.20

7
Prediction of In-situ Elastic Properties and Pore Sizes
using Conventional Well Logs
Interval of Interest
SURFACE

Wellbore : Kick-Off Point (KOP)

200 ft above
Kick-Off Point
300 ft below
Kick-Off Point

500 feet of well log data

9
10
11
12
Excess Water Production from Shales
Location of Wells

14
Interval of Interest
SURFACE

Wellbore : Kick-Off Point (KOP)

200 ft above
Kick-Off Point
300 ft below
Kick-Off Point

500 feet of well log data

15
Well Data

Well Logs

Wellbore Trajectory

Monthly Production Data


16
Machine Learning Workflow
Data Cleaning, Removal of Feature
Data Scaling and
Preprocessing, and
Outliers Transformation Selection
Feature Engineering

Feature Feature Unsupervised


Supervised Learning Reduction Extraction Learning

Kendall Tau’s Test Geophysical Signatures


HWP/LWP
+ for Excess Water
Prediction
Permutation Feature Importance Detection

17
Feature Extraction
Feature Distribution for a Cluster

Mean

Median

Variance

Kurtosis

Skewness

Root Mean-Square

Inter-Quartile Range

18
Supervised Learning Results for Water
Production Prediction

1.0

0.8

0.6

0.4

0.2

0
KNN LR SVM

19
Geomodel Compression and Novel Generation
Geomodel

2000 realizations of 15km×5km reservoir; Each geomodel contains 60K cells


21
Variational Autoencoder (VAE)

Minimize the reconstruction error + Regularize the latent representation

22
Latent Codes

23
Porosity Reconstruction

24
Permeability Reconstruction

25
Conditional Geomodel Generation

26

You might also like