Date post: | 05-Nov-2015 |
Category: |
Documents |
Upload: | ronald-chevalier |
View: | 6 times |
Download: | 0 times |
Stochastic inverse modeling under realistic prior model constraints with multiple-point geostatistics
Jef Caers
Petroleum Engineering DepartmentStanford Center for Reservoir ForecastingStanford, California, USA
AcknowledgementsI would like to acknowledge the contributions of the SCRF team, in particular Andre Journel and all graduate students who contributed to this presentation
QuoteTheory should be as simple as possible,but not simpler as possible.
Albert EINSTEIN
Overview Multiple-point geostatistics Why do we need it ? How does it work ? How do we define prior models with it ?
Data integration Integration of multiple types/scales of data Improvement on traditional Bayesian methods
Solving general inverse problems Using prior models from mp geostatistics Application to history matching
Part IMultiple-point Geostatistics
Limitations of traditional geostatisticsVariograms EW312Variograms NS2-point correlation is not enough to characterize connectivity A prior geological interpretation is required and it is NOT multi-Gaussian123
Stochastic sequential simulation Define a multi-variate (Gaussian) distribution over the random function Z(u)
Decompose the distribution as followsOr in its conditional form
Practice of sequential simulation
Multiple-point GeostatisticsReservoirmultiple-pointdata eventP ( A | B ) ?SequentialsimulationAB
Extended Normal Equationsuua
Single Normal Equation
The training image module
Training image module = standardized analog modelquantifying geo-patternsSNESIM algorithmRecognizing P(A|B) for all possible A,B
The SNESIM algorithmTraining imageData template (data search neighborhood)Search treeConstruction requires scanning training image one single time
Minimizes memory demand
Allows retrieving all training cpdfsfor the template adopted!
Probabilities from a Search Tree uSearch neighborhoodSearch treeTraining image 5 4 3 2j=1i = 1 2 3 4 5 u uLevel 0 (no CD)Level 1 (1 CD)Level 4 (4 CD)...... 1 2 3 4 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 u 1 2 3 4
Example400 sample data RealizationTrue imageTraining imageCPU
2 facies, 1 million cells
= 4 30
On 1GHz PC
Where do we get a 3D TI ?Training image requires "stationarity" Only patterns = "repeated multipoint statistics" can be reproducedValid training imageNot Valid
Modular training imageModular ? * no units * rotation-invariant * affinity-invariantTraining ImageModels generated with snesimusing the SAME training image
Properties of training imageRequired
Stationarity: patterns by definition repeat Ergodicity: to reproduce long range feature => large image Limited to 4-5 categories
Not required Univariate statistics need not be the same as actual field No conditioning to ANY data Affinity/rotation need not be the same
Part IIMultiple-point Geostatisticsand data integration
Simple question, difficult problemA geologist believes based on geological data that there is 80% chance of having a channel at location X
A geophysicist believes based on geophysical data that there is 75% chance of having a channel at location X
A petroleum engineer believes based on engineering data that there is 85% chance of having a channel at location X
What is the probability of having a channel at X ?The essential data integration problemP(A|B)P(A|C)P(A|D)P(A|B,C,D)?
Combining sources of information
Conditional independenceO = In practice not necessarilyYES
Correcting conditional independence
Permanence of ratios hypothesis
Advantages of using ratios No term P(B,C), hence McMC is not required
Work with P(A|B),P(A|C), more intuitive than P(B|A),P(C|A)
Verifies all consistency conditions by definition
It is still a form of independence, Yet dependence can be reintroduced
Simple problemP(A|B) = 0.80
P(A|C) = 0.75 => P(A|BC)
Suppose P(A) = 0.5 => P(A|BC) = 0.92Suppose P(A) = 0.3 => P(A|BC) = 0.95 = compounding of events
Lesson learned : if geologist and geophysicist agreefor almost 80%, you can be even more certain that thereis a channel !
Example reservoirP(A|C)Training imageP(A|B)Single realization
P(A|C), A = single-point !P(A|C)RealizationWhen combing P(A|B) fromgeology and P(A|C) from seismic to P(A|BC),
A is still a single point event !
Certain patterns, such aslocal rotation will be ignored
Honor seismic only as a single-point probability?
Concept of MODULAR training imageModular ? * Stationary patterns * rotation-invariant * affinity-invariant * no unitsModular Training ImageModels generated with snesimusing the SAME training image
Local rotation angle from seismicP(A|C)Local angle
Results2 realizations with anglewithout angle
Constrain to local channel featuresP(A|C)Hard data from seismicSoft data from seismic
Part IIIInverse modeling withmultiple-point geostatistics
Application to history matching
Production data does not inform geological heterogeneityaaaaaaaaaaaaaa
ApproachMethodology
Define a non-stationary Markov chain that moves a realization to match data, two properties
At each perturbation we maintain geological realism use term P(A|B)
Construct a soft data set P(A|D) such that we move the current realization as fast as possible to match the data => Optimization of the Markov chain at each step
Methodology: two faciesD = set of historic production data (pressures, flows)Some notation:Initial guess realization: Realization at iteration
Define a Markov chainDefine a transition matrix:
Transition matrix2 x 2 transition matrix describes the probability of changing facies at location u and we define it as follows
Parameter rD
Determine rDUse P(A|D) as a probability model in multiple-point geostatistics
Combine P(A|B) (from training image) with P(A|D) from production data D into P(A|B,D)
Allows generating iterations that are consistent with prior geological vision
Allows combining geological information with production data
Allows determining an optimal value for rD as follows
rD determines a perturbationrD=0.01Some initial modelrD=0.1rD=0.2rD=0.5rD=1Find rD that matches bestthe production data= one-dimensional optimization
Complete algorithm Construct a Training Image with the desired geological continuity constraint
Use snesim (P(A|B)) to generate an initial guess
Until adequate match to production data D
Define a soft data P(A|D) as function of rD
Perform snesim with P(A|D) to generate a new guess
Find the value of rD that matches best the data D
ExamplesGenerate 10 reservoir modelsthat 1. Honor the two hard data 2. Honor fractional flow 3. Have geological continuitysimilar as TIIP
Single model
rD values, single 1D optimizationrD valueObjective function
Different geology
More wells
Hierarchical matching* First choose fixed permeability per facies,perturb facies model
* Then, for a fixed facies perturb the permeability within facies(using traditional methods, ssc, gradual deformation
ExampleIP
ResultsKlow = 50Khigh = 500Klow = 50Khigh = 500Klow = 12Khigh = 729Klow = 12Khigh = 729Klow = 11Khigh = 694Klow = 150Khigh = 750
Results
More realisticReferenceInitial modelmatched model
ConclusionsWhat can multiple-point statistics provide
Large flexibility of prior models, no need for math. def.
A fast, robust sampling of the prior
A more realistic data integration approach than traditional Bayesian methods
A generic inverse solution method that honors prior information
More on conditional independenceQ? why should eB and eC be independent unless they are homoscedastic, i.e. independent of A ?Q? Is it not a mere transfer of independence hypothesis to eB and eC