Você está na página 1de 48

CMOST

CMGs Assisted History-Matching & Optimization Tool


Perth December 6, 2011

Presentation first created by CMGs Chaodong Yang, Long Nghiem, Colin Card & Rob Eastick For CMG Technical Symposium July 7-9 2010 - Calgary

CMOST Functions
CMOST can be used to perform:

Sensitivity Analysis (SA) History Matching (HM) Optimization (OP) Uncertainty Assessment (UA)

CMOST works with CMGs reservoir simulators

CMOST History Matching Field Case Study


C. Yang, L. Nghiem, C. Card, CMG M. Bremeier, Wintershall

SPE109825, 2007 SPE Annual Technical Conference and Exhibition

39

Model Overview
G4-97 G4-97 G6-97 G6-97 G15-97 G14-97 G16-97 G17-97 G2-97 G2-97 G9-97 G1-97 G19-97 G19-97 G1-97 G18-97 G9-97 G7-97 G7-97 G10-97 G10-97

0.00 0.00 ,000 551,000 553,000 555,000 557,000 559,000 561,000 563,000

1.50 2.50 565,000

3.00 miles 5.00 km 567,000

3,193,000 3,195,000 3,197,000 3,19

Permeability I, K Layer: 7
40

Model Overview
1,219,617 blocks (290,649 active) Highly heterogeneous perm distribution 55 faults with unknown transmissibilities 12 production wells with 10 years of production history All wells hydraulically fractured Wells that were difficult to match manually: G1, G3, G7, G10 IMEX used for this black oil history match

41

CMOST History Match Parameterization

Parameters selected for variation during HM

1 critical gas saturation 5 fault transmissibility multipliers 11 well productivity index multipliers 28 permeability multipliers 1 volcanic rock boundary for well G3 1 well perforation file for well G9

A total of 47 parameters Search space > 1020

42

CMOST History Matching Results


550 total IMEX runs 4 simultaneous 2-way parallel configuration Runs made on 8-core Xeon 5400 3.2 Ghz PC Total calendar time = < 3 days to reduce Global HM error from 20% to < 7%

43

Base Case Results (Well G1-97)


G1-97
Oil Rate SC (m3/day)
Time (Date)

Water Cut SC - %

44

Oil Rate SC (m3/day)

G1-97

Time (Date)

Water Cut SC - %

45

Base Case Results (Well G3-97)


G3-97
Oil Rate SC (m3/day)
Time (Date)

Water Cut SC - %

46

Oil Rate SC (m3/day)

G3-97

Time (Date)

Water Cut SC - %

47

CMOST History Match Conclusions


CMOST can handle a large number of parameters & manage a large number of runs relatively quickly

47 parameters (> 1020 combinations) 550 runs in 3 days with one 8-core 3.2 GHz Xeon 5400 PC

Successful history match workflow demonstrated

Reservoir engineer: correct parameterization and objective function definition CMOST: effective and efficient optimization algorithm Hardware: sufficient computing power

3 days versus 4 months


48

Optimization of 6-Well-Pair SAGD Model on Well locations Operating Conditions

49

Model Overview
Pair6
0.00 200 300 400 500 0.00 600 315.00 100.00 700

Pair5
630.00 feet 200.00 meters 800

Pair4
0

Pair3

Pair2

Pair1

SAGD Model Cross Section


50

-180 -190 -

CMOST SAGD Optimization


Two competing objectives

Produce more oil Reduce steam injection

Objective function

Bitumen price = $30/bbl Steam cost = $6/bbl Interest rate = 10% yearly Capital = $5,000,000 per well pair
51

NPVfield = NPVW1 + NPVW2 + NPVW3 + NPVW4 + NPVW5 + NPVW6

Base Case Results


Base case

Steam injection pressure: 2600 kPa Steam injection temperature: 226 C


Summary of base case simulation results

Steam injected (Mm3) Oil produced (Mm3) Steam-oil ratio Field NPV (M$)

13.97 3.58 3.90 119

52

CMOST Optimization Parameters


Well location

Well depth (16 parameters) Well horizontal location (16 parameters) Injector max steam pressure (46 parameters)
4 time intervals Steam temperature needs to be varied accordingly

Well operating conditions

Producer max steam rate (16 parameters)

Total number of parameters: 42

53

CMOST SAGD Optimization Results

54

CMOST Optimal Well Locations


Pair6
0.00 200 300 400 500 0.00 600 315.00 100.00 700

Pair5
630.00 feet 200.00 meters 800

Pair4
0

Pair3

Pair2

Pair1

-180 -190 -

Base Case

Optimal Case
55

CMOST Optimal Operating Conditions Well Pair 1


Well Bottom-hole Press

Time (Date)

56

Summary of CMOST Optimization Results

Item

Base case 13.97 3.58 3.90 119 1 1 hour with 4 cores

CMOST optimization 11.89 3.99 2.98 216 565 14 days with 2 x 4 cores

Change -15% +11% -24% +81%

Steam injected (Mm3) Oil Produced (Mm3) Steam-oil ratio


Field NPV (M$) Number of runs Computer time (Xeon 5400)

57

Advanced Topics Using JScript


With dynamic JScript code execution, users can customize and extend CMOST via custom code JScript code can appear in

Parameters Objective functions Constraints and penalty functions Use Corey equation to create relative permeability tables Link to Excel spreadsheet for NPV calculation Read simulation log to obtain material balance error
58

Examples

Advanced Topics Integration


CMOST will run the user-specified executable as a final pre-processing step when building dataset

BUILDER can be run silently


This makes it possible to link CMOST with any application that supports batch processing, such as geological modelling tools like GOCAD and JewelSuite

59

Recent CMOST Applications


CO2 sequestration optimization Flex well optimization ASP history matching & optimization Dynagrid tuning Numerical tuning Robust Optimization under Geologic Uncertainty (SPE 141676)
90 100 110 120

10

60

Robust Optimization of SAGD Operations under Geological Uncertainties

Chaodong Yang, Colin Card, Long Nghiem, and Eugene Fedutenko Computer Modelling Group Ltd.

Outline
Introduction

Nominal and robust optimization Challenges of robust optimization Proposed Method

Workflow
Robust optimization workflow SAGD Performance Index Optimization algorithm

Case Study
Conclusions and Ways Forward

62

Nominal and Robust Optimization


Nominal optimization

Based on a single realization (but is it right?) Ignores geological uncertainties The validity of optimum solutions is often challenged

Robust optimization
Account for geological uncertainty Seeks an optimal risk weighted solution that is most likely to give good performance for any realization of the uncertainty

63

Challenges of Robust Optimization


Run for realization 001 Run for realization 002 One set of parameter values Run for realization 003 Calculate robust objective

Run for realization 099 Run for realization 100

For each set of parameter values, 100 simulation runs are required. So the computation cost is 100 times higher than nominal optimization.

64

Challenges of Robust Optimization


Few cases will allow 100 x increase in simulation

CMGs Solution:

Classify like realizations to reduce total geomodels to investigate (SAGD Performance Index or SPI) Improve Optimization Engine
DECE Method combined with Proxy Modelling

65

Proposed Realization Ranking


Realization R1 Realization 001 Realization 002 Realization 003 Rank all realizations with CMG SPI Pick NR realizations for robust optimization Realization R2 Realization R3 Realization R4 Realization R5 Realization R6 Realization R7 Realization R8 Realization R9
66

Realization 099 Realization 100

SAGD Performance Index


Classify like realizations

Expected productivity considering well connectivity and tortuosity

Pick a cell
Figures out a path from the injector to this cell, then from this cell to producer

Calculates the harmonic average permeability of this path

Calculate harmonic average the same path

assuming clean sand along

67

SAGD Performance Index


Connectivity equals the ratio of average permeability for the realization to the average permeability for the clean sand case

The largest connectivity of a cell path from the injector to producer

represents the optimum

Certain rules are applied in the search for optimum path

Steam tends to rise Oil tends to drain downwards Steam/Oil will take path of least resistance In order to reach the destination, the steam and oil can occasionally break these 3 rules (such as oil above shale layer may travel horizontally first)
68

SAGD Performance Index


The SAGD Performance Index is the average optimum connectivity of the cells in the realization If there are multiple well pairs, the above formula will be applied to all the well pairs

69

Optimization Using Proxy


Generate initial Latin hypercube design

Run simulations using the design Polynomial Ordinary kriging

Get initial set of training data

Build a proxy model using training data Add validated solutions to training data

Find possible optimum solutions using proxy

Run simulations using these possible solutions

No

Satisfy stop criteria? Yes Stop


70

Optimization Examples

Kriging 30 LHD Runs

Kriging 60 LHD Runs

Kriging 180 LHD Runs

DECE

71

Case Study
Where do we place our SAGD Well Pairs? How do we operate these Well Pairs? How do I ensure my design is optimal for all possible geologic scenarios?

72

Reservoir Model

73

Objective Function
NPV
Bitumen rate Steam rate Bitumen price Well capital Time interval Discount rate Steam cost

74

Optimization Parameters
Location of Each Well Pair

Depth Injector/Producer Spacing Well Pair Spacing

Injection Pressure
Can vary with Time 2400-3000 kPa

Injection Rate
Can Vary with Time

Producer Sub-Cool (steamtrap) Temperature


75

Ranking of Realizations

76

Run Progress

77

Optimal Well Operation

78

Optimal Well Placement

79

NPV Histograms

80

NPV Cumulative Probability

81

Computational Cost

9 realizations is still probably excessive, but a minimum of 3 is critical.


82

Conclusions
Robust optimization is able to find an optimal risk weighted solution that gives good performance for any plausible geologic realization To make robust optimization practical for SAGD process, we need a workflow that can

Account for the impact of geological uncertainty on optimization Significantly reduce computational time

Key technique of the workflow presented Rank the entire set of geological realizations Select a small set of representative realizations for robust
optimization

83

CMOST

Thank you for your time! Any Questions?

84

Você também pode gostar