Scribd
Upload
239 views32 pages

Advances in CMOST AI Development: Chaodong Yang

This document summarizes recent advances in CMOST AI development, including enhancements to the Bayesian engine such as automated machine learning and parallelization. It describes a workflow integrating geology and optimization tools. Integration of geomechanics data is also discussed. The document highlights ease-of-use features like stacked bar charts and cloud computing capabilities. It concludes by acknowledging CMOST developers.

Uploaded by

lvt2906
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
239 views32 pages

Advances in CMOST AI Development: Chaodong Yang

This document summarizes recent advances in CMOST AI development, including enhancements to the Bayesian engine such as automated machine learning and parallelization. It describes a workflow integrating geology and optimization tools. Integration of geomechanics data is also discussed. The document highlights ease-of-use features like stacked bar charts and cloud computing capabilities. It concludes by acknowledging CMOST developers.

Uploaded by

lvt2906
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
You are on page 1/ 32

Advances in CMOST AI Development

Chaodong Yang, Ph.D., P.Eng.


Staff Engineer – Coordinator CMOST

0
1
Outline

• Bayesian Engine Enhancements


• Geology to Optimization Workflow
• Geomechanics Integration
• Ease of Use Features
• Cloud Computing
• cEDIT Enhancements

2
Bayesian Engine Overview
HM is a non-unique problem. Multiple combinations
of input parameters can result in equally-good match.

What is Bayesian Engine?


• DECE/PSO/DE Engine
• HM as an optimization problem
• Find a place with water
• Bayesian Engine
• HM as a sampling problem
• Find all places with water

3
Importance of Multiple HM Models
Single HM Model Multiple HM Models

4
Bayesian Engine Enhancements in 2018GR
Machine Learning by Parallelization of Bayesian Engine
Multilayer Neural Network

Simulation
Management

Bayesian PAR Stack

Automated
Machine
Learning

5
Key Technologies and Benefits
• Automated Machine Learning
- Automatically suggests how many experiments to run
- Automatically designs neural network architectures
- Automatically trains neural network models
- Automatically explores uncertainty space
- Engine becomes smarter as it learns
• Parallel Bayesian Engine
- CPU-intensive machine learning runs in parallel with simulation
management tasks
- Scalable to large clusters and cloud computing
- Scalable to many simulation jobs (thousands)
6
Eagle Ford Case Study
• Eagle Ford “oil window” well, 4000 ft
- 10-stage fracture job pumped
• 7 months of production data
- Oil, gas & water rates, and flowing BHP measured daily
• 16 history match parameters
- Reservoir matrix & natural fracture
- Propped fracture
- Compaction table
- Relative permeability end points and exponents

7
Bayesian HM with Multilayer Neural Network

CMOST AI:
• Trained 4,000 neural network
models of different
architectures
• Evaluated 800,000,000
possible combinations of
reservoir and hydraulic fracture
parameters
• Created and ran 4,500 IMEX
simulation models

Total Run Time: 2 days on a machine with 20 cores


8
What’s the Value for Running 4500 Jobs?
Bayesian with multilayer NN Differential evolution
4,500 Jobs 800 Jobs

Simulations with HM Error <= 4%


9
What’s the Value for Running 4500 Jobs?
Bayesian with multilayer NN Differential evolution
4,500 Jobs 800 Jobs

Simulations with HM Error <= 4%


10
What’s the Value for Running 4500 Jobs?
Bayesian with multilayer NN Differential evolution
4,500 Jobs 800 Jobs

Simulations with HM Error <= 4%


11
Bayesian Multilayer NN vs. RBF
Bayesian with multilayer NN Bayesian with RBF
4,500 Jobs 800 Jobs

Simulations with HM Error <= 4%


12
Bayesian Multilayer NN vs. RBF
Bayesian with multilayer NN Bayesian with RBF
4,500 Jobs 800 Jobs

Simulations with HM Error <= 4%


13
Bayesian Multilayer NN vs. RBF
Bayesian with multilayer NN Bayesian with RBF
4,500 Jobs 800 Jobs

Simulations with HM Error <= 4%


14
Choose History Matching Method

Fast Accurate
Fast HM Uncertainty Uncertainty
Quantification Quantification

DECE: designed exploration controlled evolution √


PSO: Particle swarm optimization √
DE: differential evolution √
Latin hypercube plus proxy √
Bayesian with RBF (radial basis function) √
Bayesian with multilayer neural network √

15
Geology to Optimization Workflow
Builder Simulation
GOCAD Model
Petrel
New in
2017GR

CMOST
Algorithm Result
Analysis

16
CMOST – Geomechanics Integration
• Read and Analyze Geomechanics Grid Results
• Shear stress
• Normal stress
• Subsidence
• Vertical displacement
• …

17
Well and Recurrent Event Optimization
How do we
parameterize the time?

18
Plackett-Burman Design Extension
• PB design requires the least number of runs
• PB design is popular for sensitivity analysis
• Extend PB design for up to 50 parameters

200-Run Latin Hypercube Design 32-Run Plackett-Burman Design

19
Manually Run Datasets

20
Stacked Bar Charts

21
Combine Histograms from Multiple Studies

Easily Compare Different Development Plans

22
Apply Plot Properties to All Plots

23
Export All Time Series Data for Analytics Tool

Time Series
Data
CSV

24
Cloud Computing – Run Simulations Anywhere

coming
soon

25
Run Simulations on Cloud 1-2-3
Register
and log in

Select the
cloud Set No. of
concurrent
jobs and go!

26
cEDIT – Dataset Editor for Reservoir Engineers

Run Dataset

Condense
large arrays
Navigation

27
Compare Files with Option to Ignore Large Arrays

28
F1 Quick Access to Simulator Keyword Description

29
Acknowledgements
• Heng Li
• Eugene Fedutenko
• Arash Mirzabozorg
• Gordon Zhang
• Oscar Barajas
• Jing Zeng
• Cuong Dang
• Ali Karami
• Zahra Sahaf
• Ngoc Nguyen

30
Questions?

For more information:


31 CMG-TS@cmgl.ca

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy