Vous êtes sur la page 1sur 8

History Matching with

Principal Component
Analysis
MSc Thesis Project
Akshay Sharma
Industry Supervisor Tom
Dombrowsky
Imperial College Supervisor Martin
Blunt
1

Outline

Principal Component Analysis has


been used in pattern recognition
and image processing.

In recent years, it has been used


in Petroleum industry for history
matching.

The
history
matched
model
resulting for the PCA workflow,
conforms to both dynamic data as
well as geological settings.

The code has been successfully


tested on the Brugge Test Model
case with favourable results.
2

Objectives & Deliverables


Test the algorithm on a real
field model
Update the
incorporate
parameters

algorithm to
other

Resulting
in
a
history
matched model for the
field; and updated code
In the future, after testing
the code on several field
models, it can be sold as a
commercial product.
3

Contribution to the Industry

Robust and efficient workflow which results in History


matched model conforming to both historical dynamic data
and geological data.

This will lead to a more accurate foundation upon which


superior forecasts can be made.

Accurately estimate possible error in forecasting by


quantifying uncertainty associated with the history matched
model.

Capability to be made into a commercial product in the


future.

Milestones
Month 1 Update the current code from Petrel 2010 to 2011;
refactoring
with regards to objective function;
update the real field model
to Petrel 2011, sanitize
data and generate prior realizations;
First run with
the updated code.

Month 2 Based on results from first run change the algorithm


by
including more parameters; some data need to be
smooth out;
improve the objective function,
leading to a robust and
efficient code.

Month 3 Generate results after several simulation runs; start


on the
run of Brugge
PCA with EnKF.

report writing; If time permits look at the second


Test Model and possibility of tying up

Timetable
May

June

July

August

Literature Review
Update the code
Refactoring objective
funct.
1st Run of Field Model
Changes in code and
improving it
Generating Results, Report
writing

Introductio
n
Presentatio
n

Progress Reports

Report
Submissi
on

Risks & Risks Mitigation

Confidentiality issues with regards to clients model


Use of several other test models for presenting results

Problem with field model


Simulator Stability: make changes into simulator or model
Lack of Matching: add more data or sanitize the data
Time Constraint: improve the iterative part of code or use
parallel
clusters

Thank You
8

Vous aimerez peut-être aussi