You are on page 1of 8

Slot 1: Uncertainty Quantification &Analysis: Objective of Probabilistic

Forecasting
Presenter: Ralf Schulze-Riegert , Schlumberger
References:
1.
2.
3.
4.
5.

SPE 94680 (Reliability and Uncertainty in Reserves)


SPE 15358 (North Sea Score Card)
SPE 103434 (Reserves Estimation -The Challenge for the Industry)
AAPG Bulletin V81, No.7,A Process for Evaluating Exploration Prospect
Additional: Reservoir Engineering for Geologist
http://large.stanford.edu/courses/2013/ph240/zaydullin2/docs/fekete.pdf

Notes:

Uncertainty caused by:


-Limited sample
-Imperfect analogues
-Sampling bias (the norm in the oil patch)
Building complex model may not contribute to the usefulness of the
forecast
Physical distribution of uncertainty-Use log normal (porosity, NTG, API
gravity)
Knowledge distribution of uncertainty -Uniform /Triangular (OWC,GOC,etc)

Questions:
1. How important it is to mimic physics of physical distribution?
2. How you define uncertainty ranges? (P90-P10, P99-P1,P100-P0)
# Two people voice out that we should push the ranges to the limit (What
is the lowest you can get, and what is the maximum you can get)
3. When is highly complex model necessary? Do they add value to your
decision making?
4. Do your company do lookbacks? How do they use that?
# Refer to SPE papers to get the answer

Slot 2: Big Loop, Small Loop, Reservoir Forecasting


Presenter: Steve Flew
Knowledge
Making
Uncertainty
frame
-Gathering
data and
analysis

Process

Decision

Reservoir
Modelling

Production
forecast

-Static,
dynamic and
HM

-Reserve,
facilities, well
planning

Tools allow us to maintain uncertainty that are not felt during production
history
Tools that are reflect what model is learning and present information to
engineer
Challenges (feedbacks from participants):
We rarely capture the uncertainty of the tools (software)
Other disciplines did not share their knowledge regarding
uncertainty to Reservoir Engineer
What is the good history match? Are there any guidelines for us
to follow? What we need?

Slot 3: From Deterministic to Probabilistic: A fit for Purpose Modelling


Workflow to Manage Subsurface Risks and Uncertainties
Presenter: Rahimah Rahim

Why probabilistic?
Because subsurface uncertainties are unavoidable
Deterministic modelling producing P10, P50, P90 not capturing
wider range of subsurface uncertainties
Static uncertainties affect hydrocarbon:
Fluid contact
Porosity
NTG
Sand distribution
Boi, Bgi

Typical workflow for deterministic process:


Static Model
STOOIP, GIIP
P10,P50,P90
Dynamic
model

Select one static model, usually P50 case as


base model

History
Matching
Range
Adjustment
Simulation
Prediction

Deterministic
Forecast-P10, P50,
P90
Typical workflow:
Perform sensitivity -MEPO by SCHLUMBERGER
Identify heavy hitters
Perform HM -Objective function
Run prediction
Build proxy
Perform Monte Carlo on Proxy modelling
Build P10, P50, P90
Use P10, P50, P90 to assess alternatives
Economic analysis
Introduction to Hybrid Technique: Mix of deterministic &probabilistic
approach, with some customization
Greenfield: Use 3 or more simulation model represent major
uncertainty combination with different volume and connectivity
Brownfield: Single Most likely model is comparable to material
balance to estimate connected volume
Recovery affected by dynamic connected volume
Focus at PETRONAS: High Risk, High value, High Capital intensive
project

Challenge:
Can probability modelling required for all or it can be selected?
Proxy value valid on per data set used -If add GOC/OWC, forecast
will change
Use sector model to identify key uncertainties
Identify HM runs that give enough spread for forecast

Questions
Can probabilities workflow to be optimized for FDP timely?
Is Hybrid approach acceptable?
How to effectively evaluate project robustness?
Is probabilistic modelling applicable to all or only selected areas?

Slot 4: Large Uncertainty Problems: How to eat elephant?


Presenter: Dmitry Eydinov RFD
Reference: SPE 182008 (Integrated Uncertainty Quantification for Development
Planning of a Large field)

Question:
1. Can modern software do a better job in supporting uncertainties and
assisted HM workflow?
2. Can we introduce any quality data check?
3. Can we propose several standard workflow variations that could cover at
least 80% of the case?
Suggestion:
After a series of HM runs, highlight data contribute to match and suggest
for RE to have a look
Slot 5: Computer Optimization of Development Plans in Presence of
Uncertainty
Presenter: Prof John Carter, EON

Key objective is to maximise value of asset to company


Challenge: Thousands of simulations required to capture various
development scenarios with account for uncertainty
Solution :34000 simulations with different model realizations and
development scenarios were run on HPC cluster
31 models were created in order to account for uncertainty based on the
reference case
A large strain of this project was taken by the hardware. A third party
cloud solution was used which allow RE more time to concentrate
additional work.
Optimization: The final well placements have some interesting features
that challenge the normal process.

Conclusion: Easier to set up 31 models to cover uncertainty, rather than meeting


about what reference case should be like.
Slot 6: Use Predictive Analysis and 2D simulation proxy to supplement
Reservoir simulation in producing a mature field
Presenter: Buoy Rina, Leap energy
Reference: SPE 180984(Data Driven Production Forecasting Using Machine
Learning)

Neural network/
A large mature field with long history, with prediction for production can
be relied on Decline Curve Analysis (DCA)
What is Machine Learning -It gives computer ability to learn and adapt
from large amount of data and make prediction (without simulation)

Slot 7: Portfolio Analysis and Optimization. Using a stochastic approach


to quantify risk and optimize field development decision
Presenter: Gunajit Das, Halliburton
Scenario
generatio
n

Scenario
generatio
n

Technical
feasibility
(Decision
making)
Portfolio

Stochastic
model
Optimization

Facilities

Platform 1
Platform 2
Production

Mono phase
Bi phase
Multipha
Slot: Risk Analysis in EOR feasibility study

Quantitative -Hierarchy tree of parameter (concentration of


surfactant)
Qualitative -Need 1000 sample to have a representative PDF with
10s of variable
Challenge: Time
p/s: Tempest can do all except surfactant flooding
Three conditions to have good proxy:
-Good sampling -Latin Hypercube Sampling

(LHS)
-Good representation of response (NPRS)
-Good quality of runs (Confirmation runs)
EOR

Reservoir properties -Static properties, heterogeneities


Chemical properties Adsorption Langmuir isotherm, CDC
Pilot design parameter -Slug concentration (what is optimum
concentration?), slug size

Economical parameter -Operation& Chemical production price,


Revenue: Oil price

Challenge: What is the best design for EOR that will improve recovery
compared to water injection.
Conclusion: Water flooding has higher recovery compared to EOR, so we
neglected EOR approach for the field.

Slot: Improving production -forecasting for new discoveries in a mature


basin by applying Experiment of Design constrained by analogue data.

Producing analogue fields helps quantify impact of key drivers


and forecast.
Must have history from nearby fields
Example: wanted to investigate probability of carbonate
reservoir with (layer A, layer B, layer C)-percentage. In RMS,
utilize petro physics and facies model realization. In Big Loop,
use proxy model for each realization.

Tempest 7.1
Proxy model
Model (up to unlimited real.)
Realization 1: Modifier 1 ,2,3

Tempest 7.2
Stochastic Proxy
Realization 1: Modifier

1,2,3,4,5
Realization 2 :
Modifier 1,2,3
Steps:
Simulate full range of uncertainties
Establish probability curve
Select low, medium and high case
Slot: Consistently integrate static and dynamic data into reservoir
models, while accounting for uncertainty using an ensemble approach
(Ensemble Kalman Filter)

Reasons for challenges (Improper uncertainty quantification):


Over simplification /neglect uncertainties
Failure to account for uncertainty in static
(G&G)
Seismic processing
Well log (Correlation & Interpretation)
Choice of geological concept

Slot: Optimisation of decision making under uncertainty throughout


field lifetime
Reference: SPE 175121 (Bayesian Style History Matching Another Way to
Underestimate Forecast Uncertainty)

Appraisal stage:
i.
Sensitivity analysis
ii.
Find cluster model with common behaviour
iii.
Find representative for cluster

Conclusion: Minimum number of models needed to access full ranges of


uncertainty. There are many ways to identify cluster , for example classify
runs based on P10,P50,P90)