eni.com upstream and technical services Polymer injection optimization using ensemble methods Laura Dovera IOR Norway 2015 28-29 April, Stavanger
eni.com
upstream and technical services
Polymer injection optimization using ensemble methods Laura Dovera
IOR Norway 2015 28-29 April, Stavanger
upstream and technical services
§ Reservoir modelling study to implement a chemical Enhanced Oil Recovery (EOR) pilot in a giant on-shore brown field
§ quantification of EOR benefit: incremental oil and associated cash flow
§ confidence interval accounting for geological/engineering
uncertainties
§ Integrated workflow with Ensemble Kalman filter (EnKF) and Ensemble Optimization (EnOpt) § Automatic integration of production data (history match)
§ Forecast optimization of chemical injection parameters
§ Uncertainty quantification with multiple models
Introduction
upstream and technical services
The field and the pilot area
§ Giant brown field § Moderate recovery factor § Onshore sandstone
environment § Oil viscosity 3-8 cP § About 200 prod. and 40 inj. § Unfavorable mobility ratio
§ Multilayered field with 12 stacked reservoirs § Turbiditic and deltaic sandstones with intercalations of shales and evaporites § Production start-up in 1953 § Current production strategy is peripheral water injection
IOR/EOR actions Why IOR/EOR?
upstream and technical services
Polymer flooding
§ Polymer flooding is one of the selected EOR techniques polymer vs. water injection
− Higher viscosity µ polymer >> µW
− Rock adsorption i.e. Krw reduction
à Lower mobility ratio
§ Actual mobility 6.5 à expected 1-0.5, with 3-6 cP target polymer
viscosity
upstream and technical services
Project activities
§ Selection of suitable area for pilot test § Good petrophysical properties § Current low recovery factor § Possibility of implementing a dispersed injection scheme § Short producers-injectors distance
§ New core data acquisition
§ Laboratories analyses § Routine and special core analyses § Chemical study for ad hoc polymer
§ Sector model construction § Geological and dynamic model § History match § Definition and optimization of forecast strategy
wellspolymer injectorpolygon for local updating in the sector model
Legend
The pilot area
upstream and technical services
Sector model
Pilot area sector § High resolution model
§ 22 total wells
§ 14 wells in commingle
§ 6 producers, 2 injectors to match
§ Complex history match
§ 60 years production
§ Few pressure data
§ Communication with the rest of the field
§ Challenges:
§ Define multiple models to represent uncertainty
§ Automatic multiple history match with EnKF
upstream and technical services
§ The Ensemble Kalman filter is a sequential data assimilation method to update dynamic systems
§ Monte Carlo approach: representation of initial uncertainty using an ensemble of models § Geostatistics and/or parameters sampling
§ Forward step: evolution of uncertainty by forward ensemble models in time § Eclipse simulation
§ Assimilation of measurements using an error minimizing update § Update parameters and state variables
The Ensemble Kalman Filter (Evensen, 2006)
SEQUENTIAL ASSIMILATION
upstream and technical services
The EnKF idea
Water Cut
Time
Measurements with errors
Original prediction
Models prediction
Updated estimate
New models prediction
Updated estimate
New models prediction
Updated estimate
upstream and technical services
EnKF on pilot area – Ensemble generation
§ NTG
§ Porosity
§ Permeability
§ Water Saturation = fun(Porosity)
Multiple porosity grid Multiple permeability grid
• Spatial variables • Generated using geological workflow
§ Water rel. perm.
§ Faults Transmissibility
• Scalar variables • Sampled within uncertainty ranges
upstream and technical services
§ EnKF updating: realizations are adjusted adding data mismatch weighted by the Kalman Gain matrix
Localization
Permx‘ = Permx + (dobs-dsim)
Kalman gain weights for permeability in layer 5 Probable spurious
correlations
Probable true correlations
Pressure data Assimilation well I2
upstream and technical services
§ EnKF updating: realizations are adjusted adding data mismatch weighted by the Kalman Gain matrix
Localization
Kalman gain weights for permeability in layer 5
= *
Localization region Localized Kalman gain
Pressure data Assimilation well I2
upstream and technical services
EnKF on the pilot area – Summary
Multiple porosity realizations Multiple permeability realizations § Ensemble size: 100
§ Ensemble generation using 3 spatial and 6 scalar parameters
§ Match of 8 wells § Localization for each well and each measurements
§ Run EnKF for 42 assimilation steps § Rerun of the final updated realizations
upstream and technical services
EnKF on the pilot area - History match for PROD P4 PROD P4 PROD P4
PROD P4
upstream and technical services
EnKF application – Final properties update
Initial average Lnk map - layer 5 Final average Lnk map - layer 5
§ Geological coherence is preserved
§ No manual modifications
upstream and technical services
Robust (on multiple models) optimization
§ EnKF output: ensemble of 100 matched models that represent the current geological uncertainty
§ Single scenario optimization: major risk of underestimating uncertainty, even the P50 could be a very unlikely model!
§ Can we perform a robust optimization - where the target is to improve the statistical outcome of the models ?
Optimization of a single profile
Robust Optimization of a profile distribution
p90
p10
p50
upstream and technical services
Ensemble Optimization (Chen, 2008)
§ Define optimization parameters and objective § polymer concentration, rates, … § cash flow, cumulative oil, …
§ Generate an ensemble of controls and couple it with the ensemble of realizations
§ Use the ensemble to investigate the objective
§ Compute gradient with the ensemble
§ Move the ensemble towards the maximum
),( jujNPVg xy=
),,( 1 cNxx …=xeNjj
uj 1},{ =xy
upstream and technical services
Polymer flooding optimization with EnOpt
Polymer injector
§ Optimize polymer injection to increment Cash Flow § Over 10 years § For the TARGET wells
§ Optimization parameters:
§ Polymer slug concentrations (kg/sm3) § Slug cycles (time) § 3 different slugs
§ Comparison to § Natural depletion scenario § Water injection scenario
Target wells
upstream and technical services
Polymer flooding optimization – Results
Actual Development Strategy
Water Injection Strategy
Polymer Injection Strategy
Polymer injection profiles guarantee the highest oil
rates
upstream and technical services
Polymer flooding optimization – Results
P50 profiles
1st Polymer slug viscosity 3 cP
Initial water pre-flush
2nd Polymer slug viscosity 6.5 cp Driving water
injection
upstream and technical services
Polymer flooding optimization – Results
P50 profiles Polymer injection average incremental recovery equal to 16%
Water injection average incremental recovery equal to 10%
upstream and technical services
Polymer flooding optimization – Results
P50 profiles Produced water is reduced by 11% using polymer with respect to
the water injection scenario
upstream and technical services
Polymer flooding optimization – Summary
Development scenario
Additional Oil Production at 2023
(P50)
Additional technical Cash Flow at 2023
(P50)
% %
No Injection - -
Water Inj. + 10 % + 3 %
Polymer Inj. + 16 % + 8 %
§ Reference values for technical cash flow evaluation § Oil price: 90 $/bbl § Treated water cost: 1 $/bbl § Polymer cost: 4.25 $/kg § CAPEX polymer pilot: 2.0 $ million
upstream and technical services
Current status of project
§ Dispersed water injection implemented in the pilot area
§ Polymer plant commissioned
§ Start-up of polymer injection planned for Q3/2015
§ Injection strategy will be implemented according to optimization results
§ A proper monitoring plan is on going
upstream and technical services
Conclusions
§ Multiple sector models § Integrated workflow from static to dynamic
§ Multiple history matching using the EnKF § Good quality history match § No manual modifications
§ Robust optimization of polymer flooding § Polymer flooding is cost-effective § Average cumulative oil = +16% § Average Cash Flow = +8%
§ Closed loop workflow § First iteration: ensemble HM and optimization § Interesting way forward - Close the loop: update the ensemble with
most recent data and use it for continuos real time monitoring and optimization
The closed loop