E&P Decision & Risk Analysis Netherlands Institute of Applied Geoscience TNO

advertisement
D&RA for improved performance of the E&P industry
E&P Decision & Risk Analysis
by Christian Bos
Netherlands Institute of Applied Geoscience TNO
- National Geological Survey
Contents
1. Objectives of RA, tools and methods
2. Features - Events - Processes (FEP) analysis
•
Objective: HSE impact assessment
3. E&P Best Practice project
•
FUN forum for Forecasting and Uncertainty; decision-making, etc.
4. “E&P Decision & Risk Analysis”
•
•
•
•
•
•
•
Objective: improved economic performance
History, past performance E&P industry
Continuous & Discrete uncertainties
Hierarchical constrained optimization under uncertainty
Options modelling
Decision analysis
Modelling: degree of holistic processing, degree of probabilistic processing
3. Conclusions
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
2
RA objectives, methods
1. Optimizing economic performance
•
•
Internal company capital investment decision-making process
Method / tools : D&RA + similar methods
2. License application / continuation
•
•
•
External orientation on government authorities
Focus on HSE, commerciality may have to be demonstrated
Method / tools : FEP analysis, perhaps D&RA-like approaches,
monitoring methods (Value of Information in terms of DRisk)
3. Operational planning
•
•
External + internal focus: operational control
Method / tools: HAZOP
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
3
FEP methodology
Feature-Events-Processes
a scenario-based, qualitative approach
using a mental, not physical, model of FEP interrelations +
empirical evidence / expert elicitation to assess probabilities
Feature: system property
Event: (exogenous) disturbance of system equilibrium
Process: reaction of system due to disturbance
(Reaction may be subject to feedback loops, delayed,
through chain of effects, non-linear: System Dynamics)
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
4
Qualitative scenario analysis
FEP identification
FEP classification
FEP Analysis
FEP selection and interaction
Scenario definition and selection
Model concept
Model building
Conse
quenc
e
analys
is
Qualitative
Scenario
Definition
Safety Assessment Model
Development
Quantitative Impact Modelling
SA of
key factors
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
5
Risk assessment workflow (scenario approach)
Scenario analysis
Risk
/
identification
classification
1
1400
mean CO2 flux
radius
1200
2.0E-02
[kg/d/m2]
1000
1.5E-02
800
600
1.0E-02
Consequence
analysis
2.5E-02
Quantitative
Model
development
Qualitative
0.
0. Definition
Definition of
of assessment
assessment basis
basis
400
5.0E-03
storage efficiency:
90.0%
1300
1000
Risk
interaction/
grouping
Scenario
(element)
formation
Conceptual
model
development
2
Testing with
3D <> 2D
numerical
model
(natural)
analogues
Probabilistic
2D numerical
simulation
Statistical
processing/
assessment
3
3800
0.0E+00
0
200
Risk
ranking/
screening
2000
3000
4000
5000
6000
7000
time [years] since start injection
8000
9000
0
10000
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
6
1. Identification and classification of risk factors –
- Database with risk factors (FEPs)
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
7
1. Assign quantitative probability of occurrence
(expert opinion) – An example –
FEP Group
(node in relational diagram)
Probability of occurrence
in 100 years
Changes natural system
0.02
Geochemical processes &
conditions
0.086
Geomechanical human induced
0.031
Gas composition
0.036
Geomechanical, natural
0.027
Geomechanical, geochemically
0.011
Leaking seal
0.02
Leaking fault
0.01
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
8
1. Building a consistent probability framework with
Bayesian Belief Network (BBN)
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
9
2. Some FEPs may be selected for modelling fluxes
and concentrations (II); Well leakage scenario:
A realisation of CO2 saturation after 10 000 yrs
• Average values
at -300 m:
23% released
from reservoir
Maximum flux
after 1500 years
Affected area:
0.18 km2
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
10
E&P Performance
underperformance due to bias &
unwillingness to learn from past &
accept new methods
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
11
FUN - Forum for forecasting and
uncertainty evaluation (1997 – 2004)
• The Forum is an effort by the authorities and
industry in Norway to determine best practice and
methods for hydrocarbon resource and emissions
estimation, forecasting, uncertainty evaluation and
decision-making.
• 18 member companies plus Norwegian Petroleum
Directorate (NPD)
• Info www.fun-oil.org
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
12
FUN - Members
•
•
•
•
•
•
•
•
•
•
•
•
•
•
A/S Norske shell U & P
Amerada Hess Norge A/S
BP Amoco Norge UA
RWE-DEA Norge A/S
Elf petroleum Norge A/S
Enterprise oil Norge ltd.
Esso Norge AS
Idemitsu petroleum Norge a.S.
Mobil exploration Norway inc.
Fortum petroleum A/S
Norsk Agip A/S
Norsk chevron A/S
Norsk Hydro ASA
Norske Conoco AS
ENGINE Workshop 7, Leiden (C. Bos)
•
•
•
•
•
Norwegian Petroleum Directorate
Phillips Petroleum Company
Norway
Saga Petroleum a.s.
Den norske stats oljeselskap
(Statoil)
Total Norge A.S
Observers
•
•
•
Ministry of Petroleum & Energy
OLF
Miljøsok
November 8th, 2007
13
The appreciation factor in relation to
discovery volumes (NPD)
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
14
Oil Production Forecast NCS
22 fields in production
160
3
Oil production (mill Sm /år)
180
140
120
100
80
60
40
20
0
PDO - Forecast
Actual production
ENGINE Workshop 7, Leiden (C. Bos)
Fall 90
Fall 95
Fall 98
November 8th, 2007
15
Comparison of investment forecasts
for fields approved before 1997
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
16
Cost & Schedule risk
•
•
•
Schedule uncertainty usually poorly managed, incl. correlation to
costs!
Opex only treated superficially: we tend to forget implications!
Later, incremental investments not properly planned: real options,
corrective actions etc. + incremental costs not formally included.
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
17
The context:
decision-making under
quantified uncertainty
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
22
Benchmarking study (bp, ChevTex, ConPhil,
ENI, Exxon, Hydro, RWE, Statoil, Total)
FUN Benchmark study 2004
12
Integration
10
A
B
8
6
F
G
E
4
H
D
C
I
2
0
0
2
4
6
8
10
12
Probabilistic processing
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
23
Integrated uncertainty analysis helps
improving company performance
Ranking improves after introducing D&RA
Ye ar (5 ye ar pe riod e nding)
1990
0
1992
1994
1996
1998
2000
2
4
Conoco
Rank
6
Chevron
8
10
Introduction of D&RA
12
14
16
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
24
The D&RA Process, how mgt & staff
create synergy: team work!
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
25
Decision-making under uncertainty:
full life-cycle perspective
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
26
Decisions and Levels of Aggregation
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
27
Using “Risk-tolerance” as optimisation
constraint
WACC
∫
• Project Risk = IRR * pdf (IRR) d(IRR)
-∞
•
i.e. cum.prob. x average IRR, if it is <WACC
∫
0
• Project Risk = NPV * pdf (NPV) d(NPV)
-∞
•
i.e. cum.prob. x average NPV, if it is <0
• The decision-maker should then specify his/her risktolerance: for the project in question, and given other
(portfolio) considerations, which cumprob x average NPV,
i.e. if it is <0, am I prepared to accept?
• Risk-tolerance criterion can then be used as optimisation
constraint to cut out bad decision-alternatives
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
28
Multi-disciplinary data aggregation &
model integration along value chain
Aggregated data integration along value chain
Proxy
model
Proxy
model
Proxy
model
Full
model
ENGINE Workshop 7, Leiden (C. Bos)
Pdf’s of KPIs
• per activity
• per project
• per asset
• per portfolio
Tax / PSC model
(in time domain)
…… modelling
Project/asset life-cycle model
Cost Engineering
…… modelling
Conceptual Design
Geomech. / fracturing model
…… modelling
Vertical Flow Performance
Geochemical modelling
…… modelling
Geomechanical modelling
Geological modelling
…… modelling
Static
Dynamic
Well
Facilities Economic
Modelling Modelling Modelling Engineering Modelling
Seismic modelling
Multi-disciplinary data aggregation
Proxy
model
Related to
“value”:
• KPI-Targets
• Optim. criteria
• Constraints
• Risk tolerance
November 8th, 2007
29
Decision-making = value optimisation =
Optimisation constraints
Corporate / portfolio level
Asset / field level
Project level
Operational level
Optimisation criteria
hierarchical constrained optimisation under uncertainty given targets
Δvalue = Δprobability of meeting a set of predefined time-series targets at the next hierarchical
decision-level
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
30
KPIs – Key Performance Indicators to be
optimised
• Corporate, e.g.
• EPS, ROACE, ROCE, RRR, Production Income; Quality of
Earnings; Production Replacement Ratios, Excluding Acquisitions
& Divestments; Finding & Development Costs, Including
Acquisitions & Divestments; Discounted Future Net Cash Flow;
Upstream Returns
• Asset, e.g.
• NPV (EMV); IRR; UTC; P/I; POT; Dproved developed reserves;
Dexpected reserves; etc.
• Project, e.g.
• Capex minimisation within time constraint
• Appraisal, e.g.
• Value of Information (DEMV)
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
31
Optimisation constraints
• Usually, cost-related KPIs
•
UTC, Maximum exposure, POT, RRR
• To be used as hurdle rate
•
E.g. WACC as hurdle rate for IRR, zero for NPV
• In the probabilistic mind-set, a risk-tolerance criterion
should be added to act as optimisation (meta-)constraint:
•
•
E.g. “I accept a probability-weighted NPV, if it is <0, of n $MM”
Then any project with a risk > n will be rejected.
• Other constraints:
•
Manpower, opportunities, HSE, time
• Integrated business models attempt to model “constrained
KPI optimisation process”
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
32
Probability of meeting portfolio multicriteria objectives in time
Ref. SPE 68576 (Howell, Tyler): Using Portfolio Analysis
to Develop Corporate Strategy
Corporate Production Planning
Production
Target production
projects
GAP
developments
ENGINE Workshop 7, Leiden (C. Bos)
2024
2022
2020
2018
2016
2014
2012
2010
2008
2006
2004
2002
2000
Producing fields
November 8th, 2007
33
Probability of exceeding portfolio
multi-criteria constraints in time
Corporate Net Cash Flow Planning
NCF constraint
NCF
projects
developments
2024
2022
2020
2018
2016
2014
2012
2010
2008
2006
2004
2002
2000
Producing fields
Risk tolerance to be specified: acceptable probability of not-meeting hurdle rate
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
34
Portfolio time-domain feedback mechanism
to be included in asset decision-making
Verify contribution of
“optimised” asset decision
against portfolio objectives.
If necessary, override standalone asset decision.

SF1 SF2
ENGINE Workshop 7, Leiden (C. Bos)
Corporate Production Planning
Production
Target production
projects
GAP
developments
November 8th, 2007
2024
2022
2020
2018
2016
2014
2012
2010
2008
2006
2004
2002
Producing fields
2000
Objective
function &
constraints
(outside time
domain!)
35
D&RA - 5 main steps
1. Frame
the
problem
•Agree dec. crit.
•Agree decisions
•Agree scenarios
•Construct tree
•Prune tree
•Agree tree
2. Set-up
quantitat.
models
•Agree models
•Populate model
•Agree stoch.
parameter pdf’s
& scenario prob.
•Agree / est.
correlations
•Agree KPIs
•Agree risk def.
3. Generate
range of
outcomes
•Est. MC run
parameters
•Pdf’s of KPI’s
•Quantify risks
•Assess impact
on portfolio
•Est. utility fct,
risk tolerance
ENGINE Workshop 7, Leiden (C. Bos)
4. Perform
Sensitivity
Analysis
•Tornado etc
•Fine-tune
decision altern.
• Test robustness of decis:
- model input
- process par
- utility fct
- dec.sequence
•VoI, VoF, ROV
5. Apply
Decision
Criteria
•Describe
process
•Propose
optim. solution
+ impact on
portfolio
•Report
•Monitor
•Update model
November 8th, 2007
36
Decision-Making framework required for valuation
(“No impact? No value!”)
Modelling decisions and uncertainties in a combined framework
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
37
D&RA step 1: Pruning the tree (1)
• 96 end-nodes
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
38
Pruning the tree (2)
• 48 end-nodes : reduced by half
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
39
The ‘Value Loop’ (Shell)©
Data
Asset Value
Drivers &
Constraints
Decisions
& Plans
Physical
Asset
Models
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
40
A typical scenario / decision tree
Decision nodes
Chance nodes
• Decision nodes, chance nodes, end-nodes (or leaves)
• What happens in end-node?
• How is total statistical information used at decision-node?
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
41
Integrated, nested models to be run
using Monte Carlo sampling process
1 pdf for each KPI
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
42
Options modelling to capture value-upside and
mitigate value-risk: flexibility has value!
• Using automatic “triggers” in
time-series (dynamic DT)
• E.g., oil price expectation after
time-step n until end of project
• Triggers can be combined
using Boolean operators, e.g.
sample
1
2
3
4
5
6
7
8
9
10
option
year 1
continue
continue
continue
continue
continue
continue
continue
continue
continue
continue
year 2
continue
abandon
continue
continue
continue
continue
continue
continue
continue
continue
yr1
yr2
100
300
Plat form
Const r.
1000
Type 2
SE_I1
Type 3
1000
I1
-10
700
horizontal
Well
Drilling
100
vert ical
100
wait
100
abandon
0
year 3
continue
year 4
continue
year 5
continue
year 6
continue
year 7
continue
year 8
continue
year 9
continue
continue
continue
continue
continue
continue
continue
continue
continue
continue
special
wait
abandon
continue
continue
continue
continue
continue
special
wait
continue
special
continue
continue
special
continue
continue
special
continue
continue
special
continue
continue
continue
continue
continue
continue continue
continue wait
continue continue
abandon
continue
wait
continue
continue
wait
continue
ENGINE Workshop 7, Leiden (C. Bos)
100
yr3
NPV
500
-10
300
800
300
100
200
100
200
30
252
E(fut. oil price) < 15
OR
CFn, n+5<0
AND
Prodn, n+5<3000
November 8th, 2007
43
Problem framing: designing and evaluating
options in dynamic decision trees
← “mapping uncertainty space onto decision space” →
Quantify
uncertainties
←
Predict probabilistically
when & how these
uncertainties may be
resolved in time (note 1)
Design, for each
scenario, options in
response to (gradual)
unveiling of truth (note 2)
option valuation (“Value of Flexibility”)
Design, for each option,
a decision algorithm based
on (expected) state variables /
KPIs (to be applied at t=t1 …)
Calculate, for each optional
decision path in time, the NPV
(by including cost of option)
→
Discontinue / delete
any sub-optimal path
(strike pull-out option)
← project valuation and ranking (including “Value of Flexibility”) →
Calculate, for all dynamically
optimized (i.e. filtered), optional
decision paths in time, the
EMV of the full project
1.
2.
Compare this to EMV of
alternative project definitions
(with D flexibilities) & rank
Select optimal
project definition
Distinguish unveiling of endogenous vs. exogenous information
•
Endogenous (project-specific): valuation of flexibility using EMV
•
Exogenous (general market, etc): valuation of flexibility using ROV
Unveiling of new info: distinguish model input (e.g. perm.) vs. model output parameters (e.g. q oil, NCF)
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
44
The task of all stakeholders in E&P
decision making is to
• Correctly quantify, using
the available models, the
uncertainty in the KPIs,
• Reduce the associated risk
(i.e. reduce the chance of
obtaining a KPI less than a
given value),
• Grasp the associated
opportunity or upside
potential (i.e. maximise the
chance of obtaining a KPI
more than a given value) by judiciously acquiring new information
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
45
The “modelling cube”
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
46
uncertainty
Utopia:
the dream
integration
Current practice
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
47
uncertainty
Utopia:
the dream
The high degree of model
precision limits what we can
achieve in terms of holistic
and probabilistic modelling
integration
Current practice
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
48
The realistic
dream?
Gradually increase
precision (decisiondriven)
uncertainty
The utopian
dream
integration
Current practice
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
49
Discrete uncertainties
(scenario trees)
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
50
Decision node
(with risk&opp. factors)
Dead-end node
(ltd. calc. of FM)
Scenario / decision
name
Scenario chance
Chance node
(can be conditional)
Optimal decision
(branch coloured red)
End node (leaf)
here calculations in
Fast Models are done
?
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
51
Continuous uncertainties
(probability density functions)
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
52
Integrated Asset Management
1 pdf for each KPI for each end-node
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
53
Monte Carlo Simulation Methodology,
uncorrelated
Pr
Randomly
Sample
Pr
*
*
Capital Expenditure
Revenue
Calculate
Pr
Pr
*
Operating Expense
ENGINE Workshop 7, Leiden (C. Bos)
Grey area =
risk of NPV<0
0
=
*
Cash Flow
November 8th, 2007
54
Monte Carlo Simulation Methodology,
correlated parameters (here >0)
Sample
correlated
Randomly
Sample
Pr
Pr
*
*
Capital Expenditure
Revenue
Calculate
Pr
Pr
Lower risk!
*
Operating Expense
ENGINE Workshop 7, Leiden (C. Bos)
*
0
=
Cash Flow
November 8th, 2007
55
Input parameters, output values and
key performance indicators
KEY PERFORMANCE INDICATORS
Indicators
Technical
Economic
STOIIP
etc...
NPV
etc...
OUTPUT VALUES
SE
STOIIP
DE
UR
RDP
Qo(t)
#Wells
SF
Qo(t)
CAPEX
AP
Qo(t)
OPEX
CO
INPUT PARAMETERS
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
56
Integrating continuous and discontinuous
uncertainties (pdf’s & scenarios)
Sample individual KPIpdf’s and time-series at
chance nodes and
construct merged pdf
NP
VP /
I IR
R
Establish pdf of KPIs for each end-node
Correctly model scenario dependencies !
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
58
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
59
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
60
Hierarchical optimization
• Optimized project should also be optimal for the
asset’s life-cycle
• Optimized asset life-cycle should also be optimal for
the company’s portfolio
• Etc.
• Risk cannot be assessed stand-alone for a project
• Risk should be assessed in context of the portfolio
• Methods & tools required to preserve uncertainty
relationships intra- & inter-project !
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
62
DSS-Portfolio imports DSSAsset information …
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
63
… and optimises phasing of projects
using objective function and EF
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
64
Probability vs. time of meeting set of
corporate KPIs is optimised
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
65
Conclusion
• State objective of RA very clearly
•
•
•
•
•
HSE ? Are HSE-norms constraints for economic optimization?
Internal decision-making for capital allocation?
License application?
Operational planning?
Etc.
• Agree to which extent all processes can be modelled
quantitatively, and whether models can be integrated
• Can impact models be integrated with economic models?
• See modelling more as a learning environment, rather than
predictor of absolute truths
• Geosystem remains mainly poorly known!
• Updating models & risk profiles, verifying assumptions…. LEARNING!
• Design monitoring activities, design new decision options, strike options
ENGINE Workshop 7, Leiden (C. Bos)
November 8th, 2007
68
Download