smart fields consortium: optimization of oil field ... fields consortium: optimization of oil field...
TRANSCRIPT
NTNU 2014
Smart Fields Consortium: Optimization of oil field
development and operations Oleg Volkov
NTNU, Sept 18, 2014
NTNU 2014
Managed by the Department of Energy Resources Engineering
Smart Fields Consortium: Baker Hughes, BG Group, BP, Chevron, CMG, ConocoPhillips, EcoPetrol, ENI, IBM, NTNU, Petrobras, Saudi Aramco, Shell
Goal: develop efficient software tools for the optimization of oil field development and operations
http://smartfields.stanford.edu
Smart Fields Consortium
Sept 18 Optimization of oil field development and operations 2
NTNU 2014
Smart Fields
Sept 18 Optimization of oil field development and operations 3
• Decision loop consists of the production system, sensors, data aggregation, planning and control
• Optimization techniques could be deployed at any stage of the development of a field
Production System
Data
Update Detailed Model
Update Reduced Model
Optimization
Controls
Optimization
Gradient Framework
Geological Model
Controls
Production system
Production Data
Seismic Data
Field Development
NTNU 2014
Optimization of field operation
Sept 18 Optimization of oil field development and operations 4
• Production optimization • Optimal well control • Few continuous and
discrete valued optimization variables
• Optimization techniques: • Reduced-order models • Proxies • Adjoint-gradient • Data mining from downhole
gauges • Streamline models • Approximate dynamic
programming • Carbon Dioxide Sequestration
Production System
Data
Update Detailed Model
Update Reduced Model
Optimization
Controls
Optimization
Gradient Framework
Geological Model
Controls
Production system
Production Data
NTNU 2014
History matching
Sept 18 Optimization of oil field development and operations 5
Data sources scarce in time and space: historical production data, seismic data etc.
Inverse modeling Numerous
continuous and discrete valued unknown variables
Geological uncertainty: quantification and propagation
Techniques: tomographic inversion PCA+adjoint gradient and PSO
Production System
Data
Update Detailed Model
Update Reduced Model
Optimization
Controls
Optimization
Gradient Framework
Geological Model
Controls
Production system
Production Data
Seismic Data
NTNU 2014
Field development
Sept 18 Optimization of oil field development and operations 6
• Well placement • Well pattern optimization • Evolutionary algorithms • Mesh adaptive direct search • Pattern search
• Drilling decision support : well and well completion design • Bayesian networks
• Discrete and
continuous variables
• Geological uncertainty
Production System
Data
Update Detailed Model
Update Reduced Model
Optimization
Controls
Optimization
Gradient Framework
Geological Model
Controls
Production system
Production Data
Field Development
NTNU 2014
Survey from E&P companies, 136 participants Data management is stated to be a challenge by 90%
Respondents admitted that they would like to know more about 25% - how to use the data which is collected for internal
decision making and external evaluation 23% - how to establish a uniform method of data
assessment 21% - how to create a data environment in which the full
potential of big data can be harnessed
Industrial perspective
Sept 18 Optimization of oil field development and operations 7
NTNU 2014
Louis Durlofsky, Khalid Aziz, Hamdi Tchelepi, Tapan Mukerji, Marco Thiele
Oleg Volkov, Vladislav Bukshtynov adjoint-based gradient framework for reservoir simulator, constrained production optimization, history matching, re-parameterization, closed-loop
Elnur Aliyev efficient field development optimization under uncertainty using upscaled models
Mehrdad Gharib Shirangi closed-loop field development optimization Hai X. Vo geological parameterization for history matching complex
models Matthieu Rousset optimization-based framework for geological
scenario determination using parameterized training images Sumeet Trehan, Rui Jiang reduced order models, TPWL, TPWQ
Contributors
Sept 18 Optimization of oil field development and operations 8
NTNU 2014
• General Purpose Research Simulator • Automatic Differentiation-based GPRS • Optimization toolkit for AD-GPRS • SNOPT (NLP solver) • PSO-MADS global search hybrid algorithm • Field development optimization framework • SGeMS
Computational tools
Sept 18 Optimization of oil field development and operations 9
NTNU 2014
Efficient field development optimization under uncertainty using upscaled models Elnur Aliyev & Louis J. Durlofsky
Sept 18 Optimization of oil field development and operations 10
NTNU 2014 Sept 18 Optimization of oil field development and operations 11
• Field development optimization methods can be computationally costly
• Optimization under uncertainty entails evaluating performance over multiple realizations (very costly)
• Flow-based surrogate models are attractive for reducing computation
• Existing reduced-order models (POD-TPWL) still only applicable for fixed well locations
Motivation
NTNU 2014
Upscaling methodology
Sept 18 Optimization of oil field development and operations 12
Global fine grid: 15x15 Global coarse grid: 5x5
x
x
x
x
x x
x
x
Well locations defined on fine grid
NTNU 2014
Computation of upscaled terms
Sept 18 Optimization of oil field development and operations 13
x
x
x
x
Global fine grid
j- j j+
NTNU 2014
Iterative upscaling procedure
Sept 18 Optimization of oil field development and operations 14
• Initial implementation by Matthieu Rousset
• Replace negative/anomalous Tj* with analytical (geometric) average
• Solve coarse model with updated Tj*
• Use new pressure values to compute Tj*
• Replace negative/anomalous (Tj* )ν with (Tj* )ν-1
• Iterate (ν ν +1)
NTNU 2014
Field development optimization
Sept 18 Optimization of oil field development and operations 15
• Implementation by Obi Isebor • PSO-MADS global search hybrid algorithm • Components:
– Well placement (discrete/categorical variables)
– Production optimization (continuous variables)
• Nonlinear constraint handling using filter method
NTNU 2014
Optimization with model refinement
Sept 18 Optimization of oil field development and operations 16
10x10 2 sec
Level 1
Simulation time:
50x50 20 sec
Level 3
100x100 120 sec
Level 4
25x25 8 sec
Level 2
Early sub-problems faster to evaluate; later sub-problems converge quickly because initial guess is close to optimum
NTNU 2014
3D Example
Sept 18 Optimization of oil field development and operations 17
• Optimize vertical well location and completion interval in 3D model
• 30 x 30 x 6 grid, BHP-controlled wells – 4 producers (fixed at 1500 psi) – 1 injector (fixed at 6500 psi)
• 4 variables per well • 20 decision variables • Economic data:
– Oil price: $100/STB – Water inj: $5/STB – Water prod: $10/STB – Well cost: $5 million/well
NTNU 2014
3D Example: Optimization performance
Sept 18 Optimization of oil field development and operations 18
10x10x1 15x15x2 30x30x3 30x30x6 Total time, hours
Upscaling & simulation time, sec
6 9 39 111
# of function evaluations (FEs) 1741 926 461 192
Run 1 $B
Run 2 $B
Run 3 $B
Average NPV, $B STD, $B
Time spent, hours
Conventional method 1.21 1.16 1.20 1.19 0.025 3.4
Model refinement 1.21 1.19 1.20 1.20 0.010 0.49
Conventional method requires 3320 objective function evaluations
NTNU 2014
Closed-loop field development optimization Mehrdad Gharib Shirangi & Louis J. Durlofsky
Sept 18 Optimization of oil field development and operations 19
NTNU 2014
Motivation
Sept 18 Optimization of oil field development and operations 20
Drill New
Well
Optimize Well Placement
Reservoir Data
Model Updating
NTNU 2014
Closed-loop
Sept 18 Optimization of oil field development and operations 21
𝑡1
Optimization
NTNU 2014
Closed loop
Sept 18 Optimization of oil field development and operations 22
𝑡1
𝑡2
Production from Well 1
Optimization
NTNU 2014
Closed loop
Sept 18 Optimization of oil field development and operations 23
𝑡1
𝑡2
History Matching
Optimization
Production from Well 1
NTNU 2014
Closed loop
Sept 18 Optimization of oil field development and operations 24
𝑡1
𝑡2
History Matching
Optimization Optimization
Production from Well 1
NTNU 2014
Closed loop
Sept 18 Optimization of oil field development and operations 25
𝑡1
𝑡2
Production from Well 1
Prod / Inj from Wells 1 & 2
𝑡3 History Matching
Optimization Optimization
NTNU 2014
Closed loop
Sept 18 Optimization of oil field development and operations 26
𝑡1
𝑡2
Production from Well 1
Prod / Inj from Wells 1 & 2
𝑡3
History Matching
History Matching
Optimization Optimization
NTNU 2014
Closed loop
Sept 18 Optimization of oil field development and operations 27
𝑡1
𝑡2
Production from Well 1
Prod / Inj from Wells 1 & 2
𝑡3
History Matching
Optimization
History Matching
Optimization Optimization
NTNU 2014
Closed loop
Sept 18 Optimization of oil field development and operations 28
𝑡1
𝑡2…𝑡6
𝑡7
History Matching
Optimization Optimization
NTNU 2014
Closed loop
Sept 18 Optimization of oil field development and operations 29
𝑡1
𝑡2…𝑡6
𝑡7
History Matching
Optimization Optimization
NTNU 2014
Optimization problem
Sept 18 Optimization of oil field development and operations 30
Objective function for field development optimization: 𝑁𝑁𝑁 = 𝐽 = 𝑝𝑜𝑄𝑜 − 𝑐𝑤𝑤𝑄𝑤𝑤 − 𝑐𝑤𝑤𝑄𝑤𝑤 − ∑𝑐𝑤𝑤𝑤𝑤
𝐽 = 𝐽 𝑢, 𝑚𝑗
𝑤 𝑢: vector of decision parameters (number of wells, well types,
controls, locations, drilling sequence)
𝑚𝑗𝑤: 𝑗-th realization updated at time 𝑡𝑤
Robust optimization:
𝐽 ̅ =1
𝑁𝑤� 𝐽 𝑢, 𝑚𝑗
𝑤𝑁𝑒
𝑗=1
NTNU 2014
Optimization problem
Sept 18 Optimization of oil field development and operations 31
𝐽 ̅ =1
𝑁𝑤 � 𝐽 𝑢, 𝑚𝑗
𝑤 𝑁𝑒
𝑗 =1
𝑀𝑤 = 𝑚1
𝑤 , 𝑚2𝑤 … 𝑚𝑁𝑒
𝑤 : is the set of realizations updated at 𝑡𝑤
𝐽 ̅ = 𝐽 ̅ 𝑢, 𝑀𝑤
Optimal solution (at 𝑡𝑤) : 𝒖𝒊 = argma𝑥 𝐽 ̅ 𝑢, 𝑀𝑤 , using PSO-MADS (Isebor et al 2013)
Use 𝑢𝑤−1 as initial guess when optimizing at time 𝑡𝑤
NTNU 2014
History matching & Bayesian Framework
Sept 18 Optimization of oil field development and operations 32
Minimize
𝑆 𝑚 =12
𝑚 − 𝑚�𝑤𝑝𝑤𝑜𝑝𝑇
𝐶𝑀−1 𝑚 − 𝑚�𝑤𝑝𝑤𝑜𝑝
+ 12
𝑔 𝑚 − 𝑑𝑜𝑜𝑜𝑇𝐶𝐷
−1(𝑔 𝑚 − 𝑑𝑜𝑜𝑜)
𝑑𝑜𝑜𝑜: observed data (vector), BHP, phase rates 𝑔 𝑚 : predicted data (vector), BHP, phase rates 𝐶𝐷: (diagonal) covariance matrix for measurement errors
Minimizing 𝑆(𝑚) gives the maximum a posteriori (MAP) estimate
Model mismatch term (prior)
Data mismatch term (likelihood)
NTNU 2014
History matching and hard data
Sept 18 Optimization of oil field development and operations 33
Minimize
𝑆 𝑚 =12
𝑚 − 𝑚�𝑤𝑝𝑤𝑜𝑝𝑇
𝐶𝑀−1 𝑚 − 𝑚�𝑤𝑝𝑤𝑜𝑝
+12
𝑔 𝑚 − 𝑑𝑜𝑜𝑜𝑤 𝑇
𝐶𝐷,𝑤−1 𝑔 𝑚 − 𝑑𝑜𝑜𝑜
𝑤
+12
𝑚ℎ − 𝑑𝑜𝑜𝑜ℎ 𝑇
𝐶𝐷,ℎ−1 𝑚ℎ − 𝑑𝑜𝑜𝑜
ℎ
𝑑𝑜𝑜𝑜ℎ : vector of observed model parameters (hard data)
𝑚ℎ: current estimate for observed model parameters 𝐶𝐷,ℎ : (diagonal) covariance matrix for measurement errors
Model mismatch term
Production data
Hard data
NTNU 2014
Randomized maximum likelihood
Sept 18 Optimization of oil field development and operations 34
Generate 𝑛 samples from the prior pdf 𝒎𝒖𝒖~𝑁 𝑚𝑤𝑝𝑤𝑜𝑝 , 𝐶𝑀
Generate 𝑛 samples as
𝒅𝒖𝒖~𝑁 𝑑𝑜𝑜𝑜, 𝐶𝐷
Minimize 𝑛 objective functions to generate 𝑁 posterior samples using L-BFGS
𝑆 𝑚 =12
𝑚 − 𝒎𝒖𝒖𝑇𝐶𝑀
−1 𝑚 − 𝒎𝒖𝒖
+12
𝑔 𝑚 − 𝒅𝒖𝒖𝑇𝐶𝐷
−1 𝑔 𝑚 − 𝒅𝒖𝒖
NTNU 2014
Subset of representative realizations
Sept 18 Optimization of oil field development and operations 35
- History match 50 realizations - Discard realizations with 𝑆𝑁 > 10
- Run realizations with 𝑢𝑤−1 - Choose 10 representative realizations
- Optimize - Drill the next well - Collect reservoir data
0 20 40-5-4-3-2-10x 108
ranked realization index
- NP
V
Ne=10
0 10 20 30 40 50
110
1001000
10000
simulation runs
SN
(m) (Similar to
Elnur Aliyev’s treatment)
NTNU 2014
3D example: 30 x 30 x 5
Sept 18 Optimization of oil field development and operations 36
Uncertain model parameters: ln(𝑘) Drill 6 wells : 3 horizontal producers, 3 vertical injectors History match and optimize over 𝑁𝑤 = 6 realizations
Layer 1
X
Y
5 10 15 20 25 30
5
10
15
20
25
30
2
3
4
5
6
7
parameter value
well cost $ 25 million oil price $ 90 / bbl produced water $ 10 / bbl injected water $ 10 / bbl drilling lag-time 210 Days reservoir Life 2000 Days perforation cost $ 1 million /gb
NTNU 2014
Optimal NPV versus Update Steps of CLFD (𝑁𝑤 = 6)
Sept 18 Optimization of oil field development and operations 37
𝐽 𝑢𝑤 , 𝑀𝑤 : Optimal E[NPV] updated at 𝑡𝑤
𝐽(𝑢𝑤 , 𝑚𝑡𝑝𝑡𝑤 ): NPV for the true model (run the true model with 𝑢𝑤)
0 210 420 630 840 1050 12608.5
9
9.5
10
10.5x 108
Time (Days)
NP
V ($
)
J(ui , Mi )J(ui , mtrue )
Deterministic
NTNU 2014
Geological parameterization for history matching complex models Hai X. Vo & Louis Durlofsky
Sept 18 Optimization of oil field development and operations 38
NTNU 2014
Motivation
Sept 18 Optimization of oil field development and operations 39
PCA is simple and linear but gives “Gaussian-looking” models and histograms
Intent with O-PCA is to modify PCA procedure to better represent geologically-complex models
NTNU 2014
History matching problem
Sept 18 Optimization of oil field development and operations 40
𝒎𝒊𝒎𝒚
𝑺 = � 𝒅𝒐𝒐𝒐𝒎 − 𝒅𝒐𝒊𝒎
𝒎 (𝒚) 𝟐𝑵−𝟏
𝒎=𝟎
+ 𝒓𝒓𝒓𝒖𝒓𝒓𝒓𝒊𝒓𝒓𝒓𝒊𝒐𝒎 𝒓𝒓𝒓𝒎
𝒚 = 𝒌𝟏, 𝒌𝟐, … . , 𝒌𝑵𝒖
𝑻
Idea is to find a geological model 𝒚 such that prediction matches production and honors prior information
NTNU 2014
Challenges of history matching
Sept 18 Optimization of oil field development and operations 41
Real models can have millions of grid blocks
Updating permeabilities for all blocks independently is expensive and may not maintain geology
Useful to have algorithms that parameterize reservoir model as 𝒚 𝝃 and maintain geology
Can then write history matching problem as:
𝒎𝒊𝒎𝝃
𝑺 = � 𝒅𝒐𝒐𝒐𝒎 − 𝒅𝒐𝒊𝒎
𝒎 (𝒚(𝝃)) 𝟐𝑵−𝟏
𝒎=𝟎
+ 𝒓𝒓𝒓𝒖𝒓𝒓𝒓𝒊𝒓𝒓𝒓𝒊𝒐𝒎
NTNU 2014
Constructing PCA representation
Sept 18 Optimization of oil field development and operations 42
Run geostatistical algorithm (SGeMS) to create a set of N realizations: 𝒚𝟏, 𝒚𝟐, 𝒚𝑵
Construct
Conceptually, define covariance matrix as
Eigen-decomposition of 𝑪 to give: (in practice use SVD of 𝒀)
K-L (PCA) representation
𝑪 =𝟏𝑵
𝒀𝒀𝑻 𝑼𝜦𝑼𝑻= C
𝒚𝒎𝒓𝒏 = 𝑼𝒓𝜦𝒓𝟏/𝟐𝝃 + 𝒚�
= 𝚽𝒓 𝝃 + 𝒚� (l << N)
𝒀 = 𝒚𝟏 − 𝒚�, 𝒚𝟐 −𝒚�, … . , 𝒚𝑵 −𝒚�
= +
𝒚𝒎𝒓𝒏 = 𝚽𝒓 𝝃 + 𝒚�
NTNU 2014
History matching using PCA
Sept 18 Optimization of oil field development and operations 43
Maximum A Posteriori (MAP) estimate
Single model
Set
Subspace Randomized Maximum Likelihood (RML):
Multiple models for uncertainty quantification
Set
𝑺 = 𝟏𝟐
𝑷𝒐𝒊𝒎 𝝃 − 𝑷𝒐𝒐𝒐∗ 𝑻 𝑪𝑫
−𝟏 𝑷𝒐𝒊𝒎 𝝃 − 𝑷𝒐𝒐𝒐∗ + 𝟏
𝟐𝝃 − 𝝃∗ 𝑻(𝝃 − 𝝃∗)
𝝃∗ = 𝝃𝒖𝒖
𝝃∗ = 𝝃� = 𝟎
NTNU 2014
Gaussian fields
Sept 18 Optimization of oil field development and operations 44
?
SGeMS realization 𝒚𝒎𝒓𝒏 = 𝚽𝒓 𝝃 + 𝒚� PCA :
NTNU 2014
Binary facies model
Sept 18 Optimization of oil field development and operations 45
SGeMS PCA
NTNU 2014
Three-facies model
Sept 18 Optimization of oil field development and operations 46
SGeMS PCA
NTNU 2014
Bimodal models
Sept 18 Optimization of oil field development and operations 47
SGeMS PCA
NTNU 2014
Optimization-based PCA
Sept 18 Optimization of oil field development and operations 48
Standard PCA: 𝒚𝒎𝒓𝒏 = 𝚽𝒓𝝃 + 𝒚�
𝒚𝒎𝒓𝒏 = 𝐚𝐚𝐚𝐚𝐚𝐚𝒓
𝚽𝒓𝝃 + 𝒚� − 𝒓 𝟐𝟐
Formulate PCA as an optimization problem:
Apply regularization + bound constraints:
𝒚𝒎𝒓𝒏 = 𝐚𝐚𝐚𝐚𝐚𝐚𝒓
𝚽𝒓𝝃 + 𝒚� − 𝒓 𝟐𝟐 + γ𝑹
𝒓 ∈ [𝒓𝒓, 𝒓𝒖]
𝑹 depends on geological model, e.g. binary 𝑹 = 𝒓𝑻(𝟏 − 𝒓)
NTNU 2014
History matching and O-PCA
Sept 18 Optimization of oil field development and operations 49
Need for gradient-based history matching
Construct using:
O-PCA gives analytically
𝒅𝑺𝒅/𝒅𝝃
𝒅𝑺𝒅
𝒅𝝃=
𝒅𝑺𝒅
𝒅𝒌𝒅𝒌𝒅𝒚
𝒅𝒚𝒅𝝃
from O-PCA
from simulator
𝒌 = 𝒇 𝒚
𝒅𝒚/𝒅𝝃
NTNU 2014
Example: three-facies system
Sept 18 Optimization of oil field development and operations 50
Oil-water with AD-GPRS
2D model, 100 x 100, 𝒓 = 100
2 injectors, q=1500 m3/d, 2000 m3/d
6 producers, BHP = 100 bar
kchannel = 2000 md, klevee = 200 md, kmud = 20 md
Data: injection pressures and production rates for 1200 days
SGeMS realizations (and O-PCA models) conditioned to hard data
Hard data
SGeMS realization
Producer Injector
20 40 60 80 100
20
40
60
80
100
0
0.5
1
1.5
2
NTNU 2014
MAP estimate
Sept 18 Optimization of oil field development and operations 51
Channel Levee
Shale
I1
I2P1P2
P3P4
P5
P6
20 40 60 80 100
20
40
60
80
100
0
0.5
1
1.5
History matched
𝑺 = 𝟏𝟐
𝑷𝒐𝒊𝒎 𝝃 − 𝑷𝒐𝒐𝒐∗ 𝑻 𝑪𝑫
−𝟏 𝑷𝒐𝒊𝒎 𝝃 − 𝑷𝒐𝒐𝒐∗ + 𝟏
𝟐𝝃𝑻𝝃
Initial guess
True model
I1
I2P1P2
P3P4
P5
P6
20 40 60 80 100
20
40
60
80
100
0
0.5
1
1.5
2
I1
I2P1P2
P3P4
P5
P6
20 40 60 80 100
20
40
60
80
100
0
0.5
1
1.5
2
NTNU 2014
Field rates
Sept 18 Optimization of oil field development and operations 52
0 1000 2000 3000 40000
500
1000
1500
2000
2500
3000
Days
TrueInitialHM
Initial
True HM
period
HM
Fiel
d W
ater
Rat
e, m
3 /d
Days 0 1000 2000 3000 4000
0
1000
2000
3000
4000
Days
TrueInitialHMInitial
True HM
HM period
Fiel
d O
il R
ate,
m3 /d
Days
Field Oil Rates Field Water Rates
NTNU 2014
Example: prediction of uncertainty
Sept 18 Optimization of oil field development and operations 53
Oil-water with AD-GPRS
2D model, 60 x 60, 𝒓 = 100
2 injectors, q = 1500 m3/d
5 producers, BHP = 150 bar
ksand = 2000 md, kmud = 20 md
Data: injection pressures and production rates for 1200 days
SGeMS realizations (and O-PCA models) conditioned to hard data
Hard data
SGeMS realization
Producer Injector
20 40 60
10
20
30
40
50
60
0
0.2
0.4
0.6
0.8
1
NTNU 2014
Prior and posterior field oil rates
Sept 18 Optimization of oil field development and operations 54
Prior models conditioned only to hard data at wells, posterior models history matched to flow data
True
HM period
Fiel
d O
il R
ate,
m3 /d
Days
Prior Models
True
HM period
Fiel
d O
il R
ate,
m3 /d
Days
Posterior Models
NTNU 2014
Optimization toolkit of AD-GPRS
Oleg Volkov & Vladislav Bukshtynov
Sept 18 Optimization of oil field development and operations 55
NTNU 2014
Optimization w.r.t. continuous-valued parameters as opposed to discrete-valued categorical parameters
Production Optimization e.g. BHP, phase rates and completion transmissibility factors
History Matching e.g. permeabilities
Validation of the gradients and AD derivatives Implementation into state-of-the-art nonlinear
programming (NLP) solvers Input data consistent with Eclipse
Features
Sept 18 Optimization of oil field development and operations 56
NTNU 2014
Input file in an ordinary text file Always starts with SIMULATOR (name of the AD-GPRS
input data file)
Content of keywords OPTDIMS, OPTPARS, OPTCONS, and OPTFUNC is consistent with Eclipse
OPTOPTS and OPTTUNE altered for tuning our optimization algorithm
A new keyword OPTCHCK is introduced for the consistency tests
Compatibility with Eclipse
Sept 18 Optimization of oil field development and operations 57
NTNU 2014
History matching is the process of building one or more sets of reservoir model parameters which account for observed, measured data
Currently parameters are permeability and interblock transmissibility (multipliers)
Sample a posteriori probability density function (pdf) using randomized maximum likelihood (RML) method
Re-parameterization
principal component analysis (PCA)
parameter grouping based on geological knowledge (faults, barriers etc.) or user defined patterns
Capabilities of History Matching
Sept 18 Optimization of oil field development and operations 58
NTNU 2014
Maximizing economic value (cash flow) Cash flow = revenues − technical costs −government take
Reduced to a linear function of production and injection rates
𝑆 = � � 𝑁𝑤𝑞𝑤,𝑤
phases
𝑤
producers
𝑤
+ � � 𝐶𝑤𝑞𝑤,𝑗
phases
𝑤
injectors
𝑗
Discounting = reducing the value of money over time to reflect the return on investment that could have been made by investing the money elsewhere
Discount factor in one year 𝑆𝑑𝑤𝑜𝑑 = 𝑆 × 1 + 𝑟𝑑𝑤𝑜𝑑/100 −1
𝑟𝑑𝑤𝑜𝑑 − discount rate in % per year Net Present Value = discounted cumulative cash surplus
Production optimization problem
Sept 18 Optimization of oil field development and operations 59
𝑁𝑁𝑁 = � 𝑆𝑑𝑤𝑜𝑑 ∆𝑡end of project
𝑡=0
NTNU 2014
Production optimization example
Sept 18 Optimization of oil field development and operations 60
Net Present Value Bounds and nonlinear constraints
Producer well liquid rate < 6000 𝑚3 𝑑𝑑𝑑⁄ Injector well water rate < 12000 𝑚3 𝑑𝑑𝑑⁄ Well water cut < 95% or Field water cut < 95% Injector BHP < 450 Bar Producer BHP > 150 Bar
𝑁𝑁𝑁 = � 11.1
𝑡365�
75 $𝑜𝑜𝑤 𝑞𝑜,𝑤 − 6 $
𝑜𝑜𝑤 (𝑞𝑤,𝑤 + 𝑞𝑤,𝑤) − 1.2 $𝑀𝑜𝑑𝑀 𝑞𝑔,𝑤 ∆𝑡
NTNU 2014
History matching problem
Sept 18 Optimization of oil field development and operations 61
Objective function (least square mismatch)
model parameters, uncertainty
, covariance matrix may be approximated by
sample covariance built on prior knowledge of geological features
analytic covariance based on analysis of available variograms
NTNU 2014
OPTDIMS – numbers of iterations and number of simulations as a termination criteria
OPTFUNC – definition of the objective function FIELD mnemonics only Production optimization: FOPT, FWPT, FWIT, FGPT, FGIT
e.g. History matching: HMOP, HMGP, HMWP, HMWI, HMGI, HMPP, HMPI e.g.
Main keywords in OptADGPRS
Sept 18 Optimization of oil field development and operations 62
OPTFUNC FOPT FIELD 471.73580778240785 0.1 / FWIT FIELD -37.738864622592629 0.1 / FWGT FIELD -0.0423776000657863 0.1 //
OPTFUNC HMOP FIELD 1.0 / HMGP FIELD 1.0 / HMWP FIELD 1.0 / HMWI FIELD 1.0 / HMGI FIELD 1.0 //
NTNU 2014
OPTPARS – definition of the control variables WBHP, WOPR, WWPR, WGPR, WGIR, WWIR – well control CTRF – well transmissibility factor control e.g.
PERM – control using permeability multipliers for history matching e.g.
LOGPARS – log-scale parameterization
Control parameters in OptADGPRS
Sept 18 Optimization of oil field development and operations 63
OPTPARS WBHP F-1H 250 450 / WBHP E-3H 150 300 2 / WBHP E-3AH 150 300 5 6 / CWIN F-1H 0 0 10 0 2 / CWIN F-3H 1:4 0 10 / CWIN E-2H * 0 10 //
OPTPARS PERM * 0 1e+3 / LOGPARS //
NTNU 2014
Given desired goal and inequality constraints involving 𝑥 and/or 𝑢 find optimal 𝑢opt
where
Unified optimization problem
Sept 18 Optimization of oil field development and operations 64
find optimum of 𝐽(𝑥, 𝑢)subject to 𝑔 𝑥, 𝑢 = 0, ℎ(𝑥, 𝑢) ≤ 0
𝑥 = state variables (pressure, saturations, etc.) 𝑔 = state equations (mass balance equations) 𝐽 = objective (Net Present Value, mismatch of observables) 𝑢 = optimization variables (well controls, modeling parameters) ℎ = nonlinear constraints (phase rates, gas-oil ratio, etc.)
NTNU 2014
Define augmented Lagrangian
where 𝜆 is the Lagrange multiplier for 𝑔
First order optimality conditions require zeros of the derivatives of Lagrangian with respect to 𝜆, 𝑥, 𝑢
Karush-Kuhn-Tucker optimality conditions
Sept 18 Optimization of oil field development and operations 65
ℒ 𝑥, 𝑢, 𝜆 = 𝐽 𝑥, 𝑢 + 𝜆𝑇𝑔(𝑥, 𝑢)
𝑔 𝑥, 𝑢 = 0 state equation 𝑔𝑥
𝑇 𝑥, 𝑢 𝜆 = −𝐽𝑥(𝑥, 𝑢) adjoint equation
𝑑ℒ = 𝐽𝑡 𝑥, 𝑢 + 𝑔𝑡𝑇 𝑥, 𝑢 𝜆 ∙ 𝛿𝑢
𝛻𝑡𝐽
min𝛿𝑡
𝛻𝑡𝐽 ∙ 𝛿𝑢
NTNU 2014
Directional derivative of objective function 𝐽 for a perturbation 𝛿𝑢
Riesz identity
OPTCHCK Number of consistency checks of the adjoint gradient Frequency of the consistency checks during the optimization Perturbation value 𝜏 Tolerance for print Type of the finite difference scheme used in the consistency check:
forward, backward, central Message CONSISTENCY_TEST_FAILED in the .sim.log file
Gradient consistency check
Sept 18 Optimization of oil field development and operations 66
𝑑𝐽 𝑥, 𝑢; 𝛿𝑢 = lim𝜏⟶0
𝐽 𝑥 𝑢 + 𝜏𝛿𝑢 , 𝑢 + 𝜏𝛿𝑢 − 𝐽(𝑥, 𝑢)𝜏
𝑑𝐽 𝑥, 𝑢; 𝛿𝑢 = 𝑔𝑡𝑇 𝑥, 𝑢 𝜆𝑇 + 𝐽𝑡 𝑥, 𝑢 , 𝛿𝑢
NTNU 2014
Optimality criteria based on Karush-Kuhn-Tucker conditions
OPTTUNE Scaling factor for the objective function
reducing objective function to ∼ 1 improves convergence for certain problems
Convergence tolerance tolerance for the KKT overall violation (in SNOPT normalized by the dual variables)
Constraint tolerance tolerance for the constraint violation (in SNOPT normalized by the control values)
Optimality tolerance tolerance on the complementarity and dual infeasibility
Tuning parameters in OptADGPRS
Sept 18 Optimization of oil field development and operations 67
NTNU 2014
ADETL: computation of Jacobian & partial derivatives of objective functional with respect to states and control variables using saved states variables (hdf5)
GMRES + AMG/SAMG/PARDISO - CPR preconditioner for adjoint system (𝑔𝑥
𝑇 is transposed Jacobian)
Bounds and constraints nonlinear programming solver
Implementation in AD-GPRS
Sept 18 Optimization of oil field development and operations 68
𝑔𝑥𝑇 𝑥, 𝑢 𝜆 = −𝐽𝑥(𝑥, 𝑢)
min𝛿𝑡
𝑔𝑡𝑇 𝑥, 𝑢 𝜆 + 𝐽𝑡(𝑥, 𝑢) ∙ 𝛿𝑢
NTNU 2014
Linear solver for adjoint system
Schur-complement for perturbed residual: 𝑅𝑅 → 𝑅𝑅 − 𝑅𝐹 𝐹𝐹−1𝐹𝑅, 𝑅𝐹 → 0
for adjoint system: 𝐽𝑥𝑅 → 𝐽𝑥𝑅 − 𝐹𝑅𝑇𝐹𝐹−𝑇𝐽𝑥𝐹, 𝐹𝑅𝑇 → 0
GMRES + CPR preconditioner for 𝑅𝑅𝑇 (Drosos Kourounis)
Linear solver
Sept 18 Optimization of oil field development and operations 69
𝑔𝑥𝑇 𝑥, 𝑢 𝜆 = −𝐽𝑥(𝑥, 𝑢)
𝑅𝑅𝑇
𝑅𝐹𝑇
𝐹𝑅𝑇
𝐹𝐹𝑇
𝜆𝑅𝑇
𝜆𝐹𝑇
= −𝐽𝑥𝑅
−𝐽𝑥𝐹
NTNU 2014
OPTOPTS – solver options Optimizer: SNOPT, IPOPT Gradient evaluation method: DISCRETE_(AD,FD),
FINITE_DIFFERENCES Linear solver:
direct: SUPERLU TRANSPOSE, PARDISO TRANSPOSE iterative + CPR preconditioner:
Krylov methods: Generalized Minimal Residual Method (GMRES), Conjugate Gradient method (CGS), or BIConjugate Gradient method (BICGSTAB)
preconditioner left (@LEFT) or right (@RIGHT) solver for pressure system (CPR_PARDISO, CPR_AMG, CPR_SAMG)
Maximum number of Krylov iterations Tolerance of the relative residual
Solvers in OptADGPRS
Sept 18 Optimization of oil field development and operations 70
NTNU 2014
Nonlinear inequalities ℎ(𝑥, 𝑢) ≤ 0
NLP solvers efficient treatment for local infeasibility require the gradient of ℎ w.r.t. 𝑢
Adjoint-based gradient of nonlinear constraints computed together with objective gradient
What about non-differentiable functionals?
e.g. ℎ 𝑥, 𝑢 ≤ 0, ∀𝑡 ⟹ max𝑡
ℎ(𝑥, 𝑢) ≤ 0
Implementation of constraints
Sept 18 Optimization of oil field development and operations 71
𝑔𝑥𝑇 𝑥, 𝑢 𝜇𝑇 = −ℎ𝑥(𝑥, 𝑢)
𝛻𝑡ℎ = 𝑔𝑡𝑇 𝑥, 𝑢 𝜇 + ℎ𝑡(𝑥, 𝑢)
NTNU 2014
Softmax Function
Sept 18 Optimization of oil field development and operations 72
Non-differentiable constraints at discrete time 𝑡𝑤
Solution: smooth maximum function + AD
max𝑤
𝑞 𝑥𝑤 , 𝑢𝑤 ≤ 𝑄
max 𝑑, 𝑏 =12
𝑑 + 𝑏 + (𝑑 − 𝑏)2+𝜀2
4 4.5 5 5.5 64
4.5
5
5.5
6
softm
ax(a
,b)
a4 4.5 5 5.5 6
-0.5
0
0.5
1
1.5
deriv
ativ
e w
.r.t.
aa
softmax(a,b), ε=1softmax(a,b), ε=0.1a b=10-a
Recursive calculation for multiple 𝑞𝑤
NTNU 2014
OPTCONS – constraints to be satisfied during optimization {F,W}OPR, {F,W}WPR, {F,W}GPR, {F,W}WIR, {F,W}GIR, {F,W}LPR – field/well phase rates
{F,W}GOR – field/well gas-oil ratio {F,W}WCT – field/well water cut
e.g. for the well P1 max
𝑡 𝑞𝑔,𝑃1 ≤ 2000, ∀𝑡 ∈ 2,3,4 input regions
for three injectors I1, I2, I3 min𝑡
min(𝑞𝑤,𝐼1, 𝑞𝑤,𝐼2, 𝑞𝑤,𝐼3) ≥ 500
for field production max𝑡
𝑞𝑔,𝑃1 + 𝑞𝑔,𝑃2 + 𝑞𝑔,𝑃3 ≤ 1.0E+6
Constraints
Sept 18 Optimization of oil field development and operations 73
OPTCONS WGPR P1 < 2000.0 2 4 / WWIR * > 500.0 / FGPR FIELD < 1.0E+6 //
NTNU 2014
Evaluate objective and constraint functionals
Evaluate adjoint based gradients
Non-Linear Programming (NLP) framework
Initialize optimization
problem
Optimization Module
Optimization Parameters
Gradient Evaluator
AD-GPRS Forward
simulator
memory alloc
bounds
Control Center
call forward solver, read output data
Objectives and
constraints
Keywords
Optimization
assemble functional
assemble gradient
Input data
NLP solver
termination criteria
new set of controls
output
Data flow
Sept 18 Optimization of oil field development and operations 74
Transpose CPR solver
call adjoint solver
NTNU 2014
Initialization 1. Through input data of the simulator
2. OPTLOAD – raw data to initialize control variables of the optimization problem.
Internal order of the control variables: it starts with an array of values corresponding to the first valid control parameter in OPTPARS at all control steps, followed by an array of the second parameter at all control steps etc.
Output .sim.log – general log file .opt.pars – objective, constraint, and controls at each evaluation of objective .opt.prnt – SNOPT log file .opt.summ – SNOPT summary file .opt.log – IPOPT log file
Initialization and output
Sept 18 Optimization of oil field development and operations 75
NTNU 2014
Examples and applications
Production optimization with economic limits HM and PCA-based parameterization History matching of geological features
HM with multiple data types Closed-loop field management
Sept 18 Optimization of oil field development and operations 76
NTNU 2014
PO with economic limits
Sept 18 Optimization of oil field development and operations 77
NTNU 2014
Previous work: maximal well flow rate constraints Kourounis et al. (2014) Adjoint formulation and constraint handling for
gradient-based optimization of compositional reservoir flow, Computational Geosciences
Volkov O., Voskov D. (2013) Advanced Strategies of Forward Simulation for Adjoint-Based Optimization, SPE163592
Conclusion: production optimization with well flow rate constraints in optimization does not perform as well as production optimization with constraints in simulation Sequential Quadratic Programming Lumped constraint using smoothed maximum function
Economic limits Well water cut limit Minimal well injection rate limit
Production optimization with economic limits
Optimization of oil field development and operations 78 Sept 18
NTNU 2014
Constraints in simulation Shut-in well
Violation of the limit is verified after each time step. If violated, the well is closed until the end of the simulation.
Stopped well Violation of the limit is verified within each time step. If violated, the well is switched to a zero-rate control, i.e. it has cross flow and can be reopened.
Constraints in optimization One lumped constraint for all wells Individual constraint for each well
Implementation of economic limits
Optimization of oil field development and operations 79 Sept 18
NTNU 2014
Van Essen et al. (2006) Robust waterflooding optimization of multiple geological scenarios, SPE102913-PA
Systematic comparison: simulations with constant time step of 30 days
Test case #1
Optimization of oil field development and operations 80
𝑁𝑁𝑁 = � 11.1
𝑡365�
75 $𝑜𝑜𝑤 𝑞𝑜,𝑤 − 4 $
𝑜𝑜𝑤 (𝑞𝑤,𝑤 + 𝑞𝑤,𝑤) 𝑑𝑡
Sept 18
BHP controlled production (4 wells) and injection (8 wells)
10 control steps of 180 days each ≈ 5 years
Economic limit: well water cut < 95%
NTNU 2014
Maximum water cut ≠ limit 0.9220 (lumped) 0.9421 (individual well) 0.9482 (shut-in well) 0.9243 (stopped well)
Number of simulations 35 (lumped) 20 (individual well) 35 (shut-in well) 33 (stopped well)
Economic limits in SQP
Optimization of oil field development and operations 81
Sequential Quadratic Programming Termination when solution cannot be improved • 20 major iterations, gradient tolerance of 10−3, feasibility region
Large major step limit, derivative-based line search
Sept 18
0 500 1000 15000
0.5
1
1.5
2
2.5
x 108
days
NP
V,$
constraint in optimization (lumped)constraint in optimization (individual well)constraint in simulation (shut-in well)constraint in simulation (stopped well)base
NTNU 2014
Economic limits in MMA and IPM
Optimization of oil field development and operations 82
Method of Moving Asymptotes Nonlinear approximation of the constraints
Interior Point Method Barrier functions for nonlinear constraints
Sept 18
Max water cut ≠ limit Number of simulations
lumped 0.9392 126 individual well 0.9479 63
shut-in well 0.8461 24
stopped well 0.9070 41
Max water cut ≠ limit Number of simulations
lumped 0.9522 91 individual well 0.9127 26
shut-in well 0.9293 1068
stopped well 0.9094 143
0 500 1000 15000
0.5
1
1.5
2
2.5
x 108
days
NP
V,$
constraint in optimization (lumped)constraint in optimization (individual well)constraint in simulation (shut-in well)constraint in simulation (stopped well)base
0 500 1000 15000
0.5
1
1.5
2
2.5
x 108
days
NP
V,$
constraint in optimization (lumped)constraint in optimization (individual well)constraint in simulation (shut-in well)constraint in simulation (stopped well)base
NTNU 2014
SQP (solution for well BHP)
MMA (solution for well BHP)
Optimal controls
Optimization of oil field development and operations 83 Sept 18
base
0 900 1800250
260
270
280
time, days
0 900 1800240
250
260
stopped well
0 900 1800250
260
270
280
time, days
0 900 1800240
250
260
shut-in well
0 900 1800250
260
270
280
time, days
0 900 1800240
250
260
individual well
0 900 1800250
260
270
280
time, days
0 900 1800240
250
260
inje
ctor
s
lumped
0 900 1800250
260
270
280
prod
ucer
s
time, days
0 900 1800240
250
260
base
0 900 1800250
260
270
280
time, days
0 900 1800240
250
260
stopped well
0 900 1800250
260
270
280
time, days
0 900 1800240
250
260
shut-in well
0 900 1800160180200220240260280
time, days
0 900 1800160180200220240260280
individual well
0 900 1800250
260
270
280
time, days
0 900 1800240
250
260
inje
ctor
s
lumped
0 900 1800250
260
270
280
prod
ucer
s
time, days
0 900 1800240
250
260
NTNU 2014
Norne Field: offshore Norwegian Sea, operated by Statoil, Norway 42,239 active grid blocks [46 × 112 × 22] Corner Point Grid, faults, and pinch-outs Real field geological and initial simulation data Base case conditioned at Jan 1, 2007 (BHP controls, 31 wells)
Test case #2
Optimization of oil field development and operations 84
red - 𝑆𝑔𝑔𝑜
green - 𝑆𝑜𝑤𝑤
blue - 𝑆𝑤𝑔𝑡𝑤𝑝
Sept 18
NTNU 2014
Production optimization outline
Optimization of oil field development and operations 85
Net Present Value Constraints
Producer well liquid rate < 6000 𝑚3 𝑑𝑑𝑑⁄ Injector well water rate < 12000 𝑚3 𝑑𝑑𝑑⁄ Well water cut < 95% Injector BHP < 450 Bar Producer BHP > 150 Bar
Optimization strategies (control variable: well BHP)
Sequential Quadratic Programming (SNOPT) Method of Moving Asymptotes (NLOPT)
𝑁𝑁𝑁 = � 11.1
𝑡365�
75 $𝑜𝑜𝑤 𝑞𝑜,𝑤 − 6 $
𝑜𝑜𝑤 (𝑞𝑤,𝑤 + 𝑞𝑤,𝑤) − 1.2 $𝑀𝑜𝑑𝑀 𝑞𝑔,𝑤 𝑑𝑡
Sept 18
NTNU 2014
MMA vs. SQP
Optimization of oil field development and operations 86
base field case MMA SQP
'06 '08 '10 '12 '14 '160
1
2
3
4x 10
8
years
wat
er in
ject
ion,
bbl
'06 '08 '10 '12 '14 '160
0.5
1
1.5
2
2.5x 10
8
years
wat
er p
rodu
ctio
n, b
bl
'06 '08 '10 '12 '14 '160
2
4
6
8x 10
7
years
oil p
rodu
ctio
n, b
bl
'06 '08 '10 '12 '14 '160
0.5
1
1.5
2x 10
9
years
NP
V, $
Economic limit: well water cut < 95% Implementation: lumped nonlinear constraint in optimization
Sept 18
10 20 30 40 50 60 70 801
2x 10
9
NP
V
objective evaluations
MMA
10 20 30 40 50 60 70 80
0.9550.96
0.97
wat
er c
ut
10 20 30 40 50 60 70 80 90 100 110
1.21.31.41.51.61.7
x 109
NP
V
objective evaluations
SQP
0.9550.96
0.97
wat
er c
ut
NTNU 2014
HM and PCA-based parameterization
Sept 18 Optimization of oil field development and operations 87
NTNU 2014
Model: 28 × 30, 3-phase black-oil [Shirangi, 2011] Synthetic history: phase rates 10 × 10 days 400 permeability realizations used to provide PCA of
transmissibility 1,622 cell transmissibilities to reconstruct Initial guess: const (mean over field)
History matching: 2D Example
5 10 15 20 25
5
10
15
20
25
30
0
1
2
3
4
5
6
5 10 15 20 25
5
10
15
20
25
0
1
2
3
4
5
6
Sept 18 Optimization of oil field development and operations 88
NTNU 2014
PCA transform uses 78 components (95% of total variance)
Main geological features captured
PCA-based Parameterization
5 10 15 20 25
5
10
15
20
25
30
0
1
2
3
4
5
6
5 10 15 20 25
5
10
15
20
25
0
1
2
3
4
5
6
5 10 15 20 25
5
10
15
20
25
30
0
1
2
3
4
5
6
5 10 15 20 25
5
10
15
20
25
0
1
2
3
4
5
6
0 100 200 300 400 500102
104
106
108
gradient evaluations
obje
ctiv
e fu
nctio
n
Sept 18 Optimization of oil field development and operations 89
NTNU 2014
HM of geological features
Sept 18 Optimization of oil field development and operations 90
NTNU 2014
Challenges
Large size of TRAN maps for 3D models
Existence of geological features (faults, barriers, etc.) ⇒ complicates the use of specific models and correspondent techniques, e.g. Gaussian
Lack of available realizations
Uncertainty in location and/or properties of geological features ⇒ available realizations in general may not be geologically consistent
History matching of real fields
Sept 18 Optimization of oil field development and operations 91
NTNU 2014
Grouping transmissibilities given fault geometry 91 controls ⇒ 8 new parameters (5 faults + 3 intersections) as
multipliers to fault transmissibilities (ranging from 0.0001 to 10) History window shrunk to 10 days (16 data points out of 165) Initial guess is 1.0 (no fault)
Geology Properties: Faults
5 10 15 20 25
5
10
15
20
25
30
0
1
2
3
4
5
6
fault geometry
0.1
0.01
0.01
0.01
10
0.001
0.0010.0001
0 5 10 15 20 250
5
10
15
20
25
30
Sept 18 Optimization of oil field development and operations 92
NTNU 2014
Faults: Reconstruction
Global minimum found short HM window no water breakthroughs “bad” initial guess
5 10 15 20 25
5
10
15
20
25
30
0
1
2
3
4
5
6
5 10 15 20 25
5
10
15
20
25
0
1
2
3
4
5
6
0 10 20 30 40 50 60 70 8010
-20
10-15
10-10
10-5
100
105
gradient evaluations
obje
ctiv
e fu
nctio
n
Sept 18 Optimization of oil field development and operations 93
NTNU 2014
Model: 16 layers from 5th to 20th History: synthetic based on real
well schedule at the beginning of 2007 (phase rates)
Norne Field: faults
faults well perforation (current layer) well perforation (other layers) active cells
Layer #5
0 5 10 15 20 25 30 35 40 45 500
20
40
60
80
100
Sept 18 Optimization of oil field development and operations 94
NTNU 2014
Example: barrier transmissibility (multiplier) history matching
8 barriers chosen manually 1,825 controls ⇒ 8 new
parameters, multipliers to barrier transmissibilities (ranging from 0.00003 to 0.25)
Initial guess: constant value 0.001
History: synthetic based on real well schedule during January 2007 (36 data points, phase rates)
Norne field: Barriers Layer #8
0 5 10 15 20 25 30 35 40 45 500
20
40
60
80
100
Sept 18 Optimization of oil field development and operations 95
NTNU 2014
Norne field: Barriers Layer #10
0 5 10 15 20 25 30 35 40 45 500
20
40
60
80
100
Layer #15
0 5 10 15 20 25 30 35 40 45 500
20
40
60
80
100
Sept 18 Optimization of oil field development and operations 96
NTNU 2014
Barrier results Barrier Name Size μ (real) μ (opt) error, % (opt)
Layer_8 168 0.02 0.02003 + 0.14 MZ_layer_10_C-segm_mid 60 0.05 0.04919 – 1.63 MZ_layer_10_C-segm_middle 588 0.25 0.24954 – 0.19 MZ_layer_10_E1 125 0.05 0.05005 + 0.09 MZ_layer_15_C_south 264 0.00003 0.0000301 + 0.46 MZ_layer_15_C_middle 421 0.00005 0.0000497 – 0.57 MZ_layer_15_E-1H 54 0.005 0.00482 – 3.62 MZ_layer_15_D-segm 145 0.01 0.01001 + 0.12
1 825 max | err | = 3.62 %
0 10 20 30
100
102
104
106
gradient evaluations
obje
ctiv
e fu
nctio
n
0 5 10 15 20 250
0.5
1
1.5
2
2.5
3
3.5x 104
accu
mul
ated
gra
dien
t
all barriers
all barriersselected barriers
Sept 18 Optimization of oil field development and operations 97
NTNU 2014
HM with multiple data types
Sept 18 Optimization of oil field development and operations 98
NTNU 2014
HM with multiple data types
Sept 18 Optimization of oil field development and operations 99
HM objective function for different types of data
production phase rates
well bottom-hole pressure
phase saturation (proxy from seismic
observations)
Production (dynamic) data: limited amount + local sensitivity Seismic data: large size + phase -dependent sensitivity How to scale , and to enhance efficiency?
NTNU 2014
Multiobjective HM
Sept 18 Optimization of oil field development and operations 100
How to scale objective coefficients to enhance efficiency?
Normalizing by phase relative density (changing volumetric rates to mass rates)
Equalizing initial contributions to the objective
Analysis by Pareto efficiency (optimality)
Objective scalarization by Multicriteria Decision Making Scheme (undergoing research)
NTNU 2014
Production vs. Seismic data
Sept 18 Optimization of oil field development and operations 101
28 × 30 black-oil 3-phase model [Shirangi, 2011]
Initial saturation Data: production 230 pts,
seismic 3,360 pts 127 PCA components
(build on 400 realizations)
5 10 15 20 25
5
10
15
20
25
30
0
1
2
3
4
5
6
7
8
9
10
5 10 15 20 25
5
10
15
20
25
30
0
1
2
3
4
5
6
7
8
9
10
5 10 15 20 25
5
10
15
20
25
30
0
1
2
3
4
5
6
7
8
9
10
5 10 15 20 25
5
10
15
20
25
30
0
1
2
3
4
5
6
7
8
9
10
0 50 100 150 20010-1
100
101
102
103
iterations
production dataseismic dataprod & seismic
NTNU 2014
Pareto optimality
Sept 18 Optimization of oil field development and operations 102
Convex hull built up only if convergence to Pareto front is guaranteed
Property of each type of data should be taken into account
Gradient-based optimization requires efficient multiple data assimilation techniques based on data structure analysis
production part,
seis
mic
par
t,
Pareto front points 10
410
5
100
101
NTNU 2014
Closed-loop reservoir management
Sept 18 Optimization of oil field development and operations 103
NTNU 2014
Reservoir management
Sept 18 Optimization of oil field development and operations 104
Production System
Data
Update Detailed Model
Update Reduced Model
Optimization
Controls
Optimization
Gradient Framework
Geological Model
Controls
Production system
Production Data
Seismic Data
• Reducing the uncertainty of the reservoir model and taking decision regarding long-term or short-term reservoir management
NTNU 2014
Example: Brugge field
Sept 18 Optimization of oil field development and operations 105
synthetic reservoir by TNO [Peters et. al., 2010] 44,550 active gridblocks (139 × 48 × 9) 10 injectors & 20 producers (smartwells) 104 geological realizations available
NTNU 2014
Reservoir management
Sept 18 Optimization of oil field development and operations 106
Reservoir 6 layers from 3rd to 8th, 2-phase dead-oil model Wells BHP controlled 10 injectors (water) & 20 producers True model realization #73
Target optimal production schedule based on updated reservoir
Production Optimization History Matching Production cycle: ~ 10 years (3,420 days)
549 BHPs (every 180 days)
Objective: NPV – oil 75 $/bbl, water 1$/bbl (inj/prod), discount 10%
History: phase rates & saturation based on and current
Initial guess : realization #101
82,582 TRAN multipliers
95 PCA components built on 104 mesh-consistent TRAN realizations
NTNU 2014
CLRM: Convergence
Sept 18 Optimization of oil field development and operations 107
2 4 6 8 100.85
0.9
0.95
1
1.05
1.1
1.15x 10
10
closed loop iteration, n
NP
V, $
4 6 8 101.148
1.15
1.152x 10
10
Optimizer: SNOPT (PO & HM) Termination: max 20 gradient evaluations + optimality conditions Performance: @ 5th year
NTNU 2014
PO: Robustness
Sept 18 Optimization of oil field development and operations 108
4 5 6 7 8 9 101.02
1.04
1.06
1.08
1.1
1.12
1.14
1.16x 10
10
years
NP
V, $
0 2 4 6 8 100
2
4
6
8
10
12x 10
9
years
NP
V, $
Notation: optimal (reference) schedule assuming known
optimal schedules using different initial guesses
NTNU 2014
Sweep efficiency: layer 5
Sept 18 Optimization of oil field development and operations 109
25 30 35 40 4530
40
50
60
70
80
90
100
110
25 30 35 40 4530
40
50
60
70
80
90
100
110
25 30 35 40 4530
40
50
60
70
80
90
100
110
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
NTNU 2014
Questions? Thank you
Sept 18 Optimization of oil field development and operations 110