Вы находитесь на странице: 1из 8

Tutorial on

Uncertainty Definition
“Ranges are more important than distributions”

Prepared by Ralf Schulze-Riegert


Schlumberger Private
MEPOversion 4.2.1
Version 03
Last Update 2 July 2016
MEPO Uncertainty Definition Page i

TABLE OF CONTENTS
Page

1 INTRODUCTION 1
1.1 Scope 1
1.2 References 1

2 CASE DESCRIPTION 2
2.1 Exercise 1 2

3 PROJECT SETUP 3
3.1 Exercise 1: 3

4 EXERCISE 1: UNCERTAINTY DEFINITION 4


4.1 Initial Uncertainty Definition 4
4.2 Modify Shapes and Ranges 5

Tutorial3_UncertaintyDefinition_160701_03_rsr.docx Schlumberger private


MEPO Uncertainty Definition Page 1

1 INTRODUCTION
This tutorial gives an introduction to parameter definitions and distribution functions in
MEPO. An introduction to the uncertainty definition is given on the basis of a synthetic test
function, STOIIP(Φ, SW, ResVol, Boi).

1.1 Scope
In this tutorial you will learn about the following MEPO project tasks:
 Exercise 1 (“Ranges are more important than shapes”)
o Introduction to distribution functions available in MEPO
o Review a prepared setup of uncertainty definitions
o Extend the project setup to show that ranges are more important compared
to the shape of a distribution function

1.2 References
[1] MEPO User Manual 2013

Tutorial3_UncertaintyDefinition_160701_03_rsr.docx Schlumberger private


MEPO Uncertainty Definition Page 2

2 CASE DESCRIPTION
This chapter gives a brief description and motivation of the simulation cases used for the
exercises.

2.1 Exercise 1
This exercise uses a synthetic test function for simulating stock tank oil initially in place
(STOIIP). The function is given by
 (1  SW ) Ah
STOIIP 
Boi
with the following abbreviations:
Ah : net reservoir volume
Boi : initial oil formation volume factor
SW : initial water saturation
φ : porosity
Distribution functions are defined for every uncertainty parameter. STOIIP is calculated
outside of MEPO using a python script which is linked to the post processor, cf. python
tutorial for details. For every experiment the result of STOIIP is written back to the global
objective function value.

Once distribution functions are defined for all uncertainty parameters, MEPO is launched
using the Latin Hypercube sampling. Simulations are repeated for different setups of the
distribution functions to investigate the statement:
“Ranges of a distribution are generally more important than shapes”

Tutorial3_UncertaintyDefinition_160701_03_rsr.docx Schlumberger private


MEPO Uncertainty Definition Page 3

3 PROJECT SETUP
All tutorials are generally built on prepared MEPO projects. MEPO project directories for
all exercises are found under C:\MEPO\. The following project directory structure is used:

3.1 Exercise 1:
Project Directory Description
C:\MEPO\TUTORIAL3\
MEPO project directory
bin Location of the python script
including the objective function
calculation (STOIIP)
TEST1 Initial definition of distribution
functions
TEST2 Alternative distribution functions
TEST3 Focus on changing ranges
STOIIP.MPRJ MEPO project file

Tutorial3_UncertaintyDefinition_160701_03_rsr.docx Schlumberger private


MEPO Uncertainty Definition Page 4

4 EXERCISE 1: UNCERTAINTY DEFINITION


The first exercise starts with a synthetic test model to quickly investigate the effects of
alternative distribution functions. The MEPO project is prepared. Only settings in the
uncertainty parameter panel need to be modified in order to investigates effects. Other
settings are prepared and will be valid for all simulation cycles.
In the first step, the existing cycle is reviewed. For actual modification work, copy the cycle
and continue the exercise.

4.1 Initial Uncertainty Definition


In the first step review the existing MEPO project.
Actions:
 Launch MEPO and open the MEPO project
o C:\MEPO\TUTORIAL3\STOIIP.MPRJ
 Review TEST1 Cycle
o Uncertainty parameters
 Review distribution functions for all parameters
o Response parameters are not defined.
 The GLOBAL will be calculated in the post processing step via an
external python script. No adjustments are required.
o Simulation configuration
 The following task elements are included:
 CreateDirectory and PostProcessor
 The PostProcessor requires directories to exist. A Python script is
launched for every post processing task.
o Review spread of results.
Uncertainty definition:

Tutorial3_UncertaintyDefinition_160701_03_rsr.docx Schlumberger private


MEPO Uncertainty Definition Page 5

Objective function definition:


The GLOBAL objective value is calculated outside of MEPO.
 (1  SW ) Ah
STOIIP 
Boi
Uncertainty parameters map as follows.
Parameter MEPO Parameter
φ PORO
SW SWAT
Ah RESVOL
Boi BOI

The dimension of GLOBAL=STOIIP can be interpreted as million SM3.

4.2 Modify Shapes and Ranges


In order to investigate effects from different distribution functions, the first cycle is copied
and then modified.
The following table lists the distribution functions used for uncertainty parameters in all
cycles to be investigated.

MEPO TEST1 TEST2 TEST3


Parameter
PORO TruncatedNormal TriangualGeneral TruncatedNormal
SWAT Uniform BetaGeneral Uniform
RESVOL TriangualGeneral Pert TriangualGeneral
BOI Deterministic Deterministic Deterministic

Actions:
 Copy cycle TEST1 to TEST2

 Delete all experiments in experiment list. They will be recalculated.


 Uncertainty Parameters: Go to the uncertainty parameter panel
o Modify the distribution definition and adjust settings such that the ranges
remain similar, e.g., P10 and P90 values should be similar in order to
compare the distributions.
 Launch simulations in new cycle
 Analyze and compare results.
o Go to Plotting Tool panel and display multiple histograms in one plot. Also
compare the distribution of the GLOBAL from cycle TEST1 and 2.

Tutorial3_UncertaintyDefinition_160701_03_rsr.docx Schlumberger private


MEPO Uncertainty Definition Page 6

 Copy cycle TEST1 to TEST3


 Delete all experiments in experiment list. They will be recalculated.
 Uncertainty Parameters: Go to the uncertainty parameter panel
o Modify the ranges from the distribution definitions. In general they should
be narrowed by a factor of 2. The mean should remain similar in order to
compare effects.
 Launch simulations in new cycle
 Analyze and compare results.

In order to investigate the initial statement, compare distributions for the GLOBAL
objective function value from all three test cycles, i.e., TEST1,2 and 3. One plot could look
like:

General task for this exercise:


Analyze effects of distribution functions, ranges and shapes and support the statement
“Ranges of a distribution are generally more important than shapes”

Tutorial3_UncertaintyDefinition_160701_03_rsr.docx Schlumberger private

Вам также может понравиться