Society of Petroleum EngineersSPE 2001–2002 Distinguished Lecturer Program4 July 2002

Reservoir Fluid Properties State of the Art and Outlook for

Future Development

!"#$

Dr. Muhammad Al-MarhounKing Fahd University of Petroleum & MineralsDhahran, Saudi ArabiaE-mail: marhounm@kfupm.edu.sa

OutlineOutline% Introduction% State of the Art% Determination of PVT properties%% Problems related to Problems related to PVT

&& Experimentation & CalculationsExperimentation & Calculations& Data smoothing & smoothing & Correlations

% Artificial neural networks%% PVT ReportingPVT Reporting% Conclusions

IntroductionIntroductionFluid Properties

The study of the behavior of vapor and liquid in petroleum reservoirs as a function of pressure, volume, temperature, and composition

Importance of PVT Properties& Determination of hydrocarbon reserves& Reservoir and simulation studies& Design of production facilities

State of the Art

& Graphical correlations are reduced to equations& Correlations have been improved& Fluid classification in reservoirs is defined & Laboratory analyses have been standardized & Chemical analyses of petroleum are made

available & EOS is utilized to calculate gas-liquid equilibria

Determination of PVT propertiesDetermination of PVT properties

& Laboratory measurements using:' Bottom hole sample' Recombined surface sample

& Equation of state with appropriate calibrations& Empirical correlations with appropriate range

of application& Artificial neural networks models

Problems related to Problems related to experimentationexperimentation

''Reservoir process Reservoir process presentation''Physical trends of lab dataPhysical trends of lab data

Reservoir process presentationReservoir process presentation& Lab tests do not duplicate reservoir process

&Petroleum engineers consider liberation process in reservoir approaches differential

& Liberation process around well is considered flash

&Actual process is neither flash nor differential

&A combination test may be closest to the reservoir process

Phase transition in oil reservoirPhase transition in oil reservoirZone A: above pbZone B: below pb, flashZone C: differential

AB

Well

Reservoir

C

Separator

OilGas

Typical trends of good lab dataTypical trends of good lab data

1.29

1.30

1.30

1.30

1.31

1.31

1500 2000 2500 3000 3500

Pressure

Vo

&Good experimental P-V data should follow physical trend. 'Volume decreases with P'Co decreases with P

' decreases with PdpdC o−

-1.54E-09

-1.53E-09

-1.53E-09

-1.52E-09

-1.52E-091500 2000 2500 3000 3500

Pressure

Slop

e of

Co

0.00

0.00

0.00

0.00

0.00

0.00

1500 2000 2500 3000 3500

Pressure

Co

T

o

oo p

VV

C

∂∂

−=1

Abnormal CAbnormal Coo trendtrend

1.35

1.36

1.36

1.37

1.37

1.38

1500 2000 2500 3000 3500

Pressure

Vo

&Co should decrease with pressure

-2.67E-10

-2.66E-10

-2.65E-10

-2.64E-10

-2.63E-10

-2.62E-10

-2.61E-101500 2000 2500 3000 3500

Pressure

Slop

e of

Co

0.00

0.00

0.00

0.00

0.00

0.00

0.00

1500 2000 2500 3000 3500

Pressure

Co

Abnormal CAbnormal Coo derivative trendderivative trend

1.21

1.22

1.22

1.23

1.23

1.24

1.24

1500 2000 2500 3000 3500 4000

Pressure

Vo

& should decrease with pressure

dpdC o−

-1.69E-10

-1.69E-10

-1.69E-10

-1.69E-10

-1.69E-10

-1.69E-10

-1.69E-101500 2000 2500 3000 3500 4000

Pressure

Slop

e of

Co

0.00

0.00

0.00

0.00

0.00

0.00

1500 2000 2500 3000 3500 4000

Pressure

Co

Problems related to calculationsProblems related to calculations

Adjustment of differential data of differential data as an exampleas an example

Adjustment of differential data of differential data to separator conditions to separator conditions --Why?

&Rs and Bo obtained by differential liberation are not the same as Rs and Bo obtained by flash liberation

&Oil leaving reservoir is flashed to separator, therefore Rs and Bo should be determined by a flash process

&Flash liberation does not cover whole range of interest, therefore differential data are corrected

Current adjustment methodCurrent adjustment method--PropertiesProperties!At lower pressure formation volume factor, Bo

might read a value less than 1

0.90

1.00

1.10

1.20

1.30

1.40

0 500 1000 1500 2000 2500

Pressure

Bo

Bo-typicalBo

t d

Current adjustment methodCurrent adjustment method--PropertiesProperties!At lower pressure, the solution gas-oil ratio, Rs

extrapolates to negative values.

-200.00

0.00

200.00

400.00

600.00

0 500 1000 1500 2000 2500

Pressure

R s

Rs-typicalRs

t d

Current adjustment methodCurrent adjustment method--PropertiesProperties

! Current adjustment method does not honor density atbubble point under reservoir conditions

ob

gsoob B

Rx γγγ

41018.2 −+=

0.71862650.738444γob

0.83430.8448γo

0.80240.9336γg

526526Rs

1.2891.289Bob

Flash Liberation

Adjusted Differential

Property

! The same crude under the same reservoir conditions, but different densities

Adjustment methods of oil FVF

& Current Adjustment of Bo

obd

obfodo B

BBB =

& Suggested Adjustment

( )obfodnobfo BBcBB −+=

)/()( odnobdodobd BBBBc −−=

Oil FVF

0.9

1

1.1

1.2

1.3

1.4

0 500 1000 1500 2000 2500

Pressure, psia

Oil

FVF

DifferentialCurrentSuggested

Adjustment methods of solution GOR

& Current Adjustment of Rs

obd

obfsdsbdsbfs B

BRRRR )( −−=

& Suggested Adjustment

( )sbdsbfsds RRRR =

Solution GOR

-100

0

100

200

300

400

500

600

0 500 1000 1500 2000 2500

Pressure, psia

Solu

tion

GO

RDifferentialCurrentSuggested

Adjustment methods methods of gas relative density

& Current Adjustment of γg

gdg γγ =

& Suggested Adjustment

)(1 gfgdgfg n

d γγγγ −+=−

)/()(111 −

−−=ngdgdgdgdd γγγγ

Gas relative density

0.6

0.8

1

1.2

1.4

1.6

1.8

0 500 1000 1500 2000 2500

Pressure, psia

Gas

rela

tive

dens

ity

DifferentialCurrentSuggested

Adjustment methods methods of oil relative density

& Current Adjustment of γo

odo γγ =

& Suggested Adjustment

)( ofodofo c γγγγ −+=

Oil relative density

0.832

0.834

0.836

0.838

0.84

0.842

0.844

0.846

0 500 1000 1500 2000 2500

Pressure, psia

Oil

rela

tive d

ensit

y DifferentialCurrentSuggested

Live oil relative density

0.7

0.72

0.74

0.76

0.78

0.8

0.82

0.84

0.86

0 500 1000 1500 2000 2500

Pressure, psia

Live

oil

rela

tive d

ensit

y DifferentialCurrent Suggesed

o

gsoor B

Rx γγγ

41018.2 −+=

Problems related to Problems related to Smoothing experimental dataSmoothing experimental data

SmoothingSmoothing relative total volume data relative total volume data as an exampleas an example

SmoothingSmoothing relative total volume datarelative total volume data

% To obtain P-V data, conduct a flash liberation experiment on a gas-oil mixture at a constant temperature

% Data analysis defines& volume & pressure at bubble point

& FVF above pb & total FVF below pb

% The experimental data as reported are accompanied by measurement errors. Therefore, the data are usually smoothed

YY--function propertiesfunction properties

1

2

3

4

5

6

0 1000 2000 3000 4000 5000

Pressure

Tota

l Rel

ativ

e Vo

lum

e

volumeY-fun value

&Only the experimental data at pressures below pb are utilized to obtain pb

&Bubble point volume is not corrected

&Y-Correlation with an error in the bubble point volume may yield a straight line but with the wrong pb

YY��Function plotFunction plot

1

2

3

4

5

6

0 1000 2000 3000 4000 5000

Pressure

Tota

l Rel

ativ

e V

olum

e volumecurve-1curve-2Y-fun valueYF

Smoothing Smoothing relative total volume datarelative total volume data

& Current

paavvvppp

ybbt

b21/)(

/)(+=

−

−=

& Suggested: add x-function beside y-function

paapppvvvx

bb

bob43/)(

/)(+=

−−

=

XX--Y Function plotY Function plot

1

2

3

4

5

6

0 1000 2000 3000 4000 5000

Pressure

Tota

l Rel

ativ

e V

olum

e volumecurve-1curve-2XY-CurveXY

YF 1944.5 1.2637XY 2014.2 1.262208

Problems related to Problems related to correlationscorrelations

''Correlation applicationCorrelation application''Properties of correlationsProperties of correlations''Physical trends of correlationsPhysical trends of correlations''Pitfalls of least square methodPitfalls of least square method

Correlation applicationCorrelation application

Correlations normally used to determine:Correlations normally used to determine:%% BubbleBubble--point pressure, point pressure, Pb

%% Solution gasSolution gas--oil ratios, oil ratios, RRss

%% Density of liquidsDensity of liquids%% Oil FVF, BOil FVF, Bob ob & total FVF, B& total FVF, Btt

%% Adjustment of BAdjustment of Bobob and and RRss

%% Oil compressibility, COil compressibility, Coo

%% Oil viscosity, Oil viscosity, µµoo ,, µµaa , , µµll

%% Interfacial tension, Interfacial tension, σσ

Properties of correlationsProperties of correlations

% Correlations typically match employed experimental data, with deviations less than a few percent

% When applied to other fluids, a much higher deviations are observed

% If fluids fall within the range of tested fluids, an acceptable accuracy can be expected

% Fluid composition could not be explained by gross properties

% Errors in some PVT correlations are not acceptable

Physical trends of correlationsPhysical trends of correlations

Trend tests are to check whether the performance of correlation follows physical behavior or not:

' Trend tests on predicted values

' Trend tests on errors

Correlation with two equationsCorrelation with two equations

1.250

1.275

1.300

1.325

1.350

1.375

1.400

10 20 30 40 50 60

Oil API Gravity

Oil F

VF

StandingMarhounVazquez & Beggs

&Modeling physical properties with two equations might produce non-physical trend

Correlation with nonCorrelation with non--physical constraintphysical constraint

1.2

1.25

1.3

1.35

1.4

1.45

0.4 0.6 0.8 1 1.2 1.4 1.6

Gas Relative Density (Air=1.0)

Oil F

VF

StandingMarhounVazquez & Beggs )( gapi γγ

&Restriction of correlation model gives non-physical trend

Correlation with limited dataCorrelation with limited data

500

1000

1500

2000

2500

60 110 160 210 260

RESERVOIR TEMPERATURE F

Pb, p

si

StandingVazquezMarhounDokla & Osman

&Correlation development for limited data will give a good fit, but might lead to non-physical trend

Trend Tests on Error: Effect of API On BTrend Tests on Error: Effect of API On Bobob

0

5

10

15

20

25

30

11.4<API<22(23)

22<API<30(39)

30<API<35(26)

35<API<40(56)

40<API<45(33)

45<API<59.2(20)

Oil API Gravity

Erro

r in

Bo

StandingVazquez & BeggsMarhoun

Trend Tests on Error: Effect of GRD On BTrend Tests on Error: Effect of GRD On Bobob

0

5

10

15

20

25

30

0.525 - 0.7(23)

0.7- 0.75(25)

0.75-0.8(24)

0.8-0.85(24)

0.85-0.9(22)

0.9 - 1.0 (27)

1.0-1.25(30)

1.25-1.7(21)

Gas Relative Density (Air=1.0)

Err

or in

Bo

Vazquez & BeggsStandingMarhoun

Pitfalls of least square methodPitfalls of least square methodUsed to estimate the regression coefficients in model

)( xfy =

% Basic assumption of LSM is the independent variable x is determinate, i.e. it has no error

% But x and y involve measurement errors, therefore

% Do not rely entirely on a method when its basic assumption is violated

Comparison of the �Best fit line�

Min y-error LSM

Min x & y-error

0 10 20 30

0.01

0.1

1

10

100

1000

40

Property

y

x

Pitfalls of logarithmic equivalencePitfalls of logarithmic equivalence

logarithmic equivalent used to linearize equations% Given the problem% Use the logarithmic equivalent

xnky logloglog +=

nkxy =

50.0419.03

8.022.51yx

% Apply LSM to minimize error% Compare errors Σδ2

Comparative error analysisError using logarithmic equivalent

)(log)(log givenyestimatedy −=δError using original values

)()( givenyestimatedy −=δΣδ2

(original problem)

Σδ2

(logarithmic equivalent)

nkMethod

13.90.568383.360.474 Iterative100.20.020982.0962.224LSM

Artificial neural networksArtificial neural networks''DefinitionDefinition''AdvantagesAdvantages''Problems & ChallengesProblems & Challenges

Artificial neural networksArtificial neural networks

A mathematical model that can acquire artificial intelligence. It resembles brain in two respects by( Acquiring knowledge through learning

process ( Storing knowledge through assigning

inter-neuron connection strengths known as weights

Neural network architectureNeural network architecture

INPUTHIDDEN

API

Rs

γg

T

OUTPUT

Bob

Pb

ANN AdvantagesANN Advantages

( Model function does not have to be known

( ANN learns behavior by self-tuning its parameters

( ANN has the ability to discover patterns

( ANN is fast-responding systems and provides a confident prediction

( ANN can accept more input to improve accuracy; such continuous enrichment or �knowledge� leads to more accurate predictive model

ANN Problems & ChallengesANN Problems & Challenges

% Design of ANN:

( Number of hidden layers

( Number of neurons in each hidden layer

( Learning constant to control speed of training

ANN Problems & ChallengesANN Problems & Challenges

% Generalization Vs. Over Fitting

( New training algorithms (cross validation)

( Hybrid systems (expert systems)

( Number of adjustable weights is large which is not justified unless the PVT data is huge

% Is the neural network the ultimate solution?

PVT ReportingPVT Reporting

'Typical PVT report'PVT report shortcoming'Suggested improvement

Typical PVT Report

& Sampling information& Hydrocarbon analysis of reservoir fluid& Oil compressibility& Pressure volume relationship (smoothed data)& Differential liberation & Separator tests & Hydrocarbon analysis of lab flashed gases& Liquid and gas viscosity data& Mixture density

PVT ReportPVT Report-- Shortcoming

% Reports smoothed results only

% Does not include raw data

% Does not verify data consistency

PVT Report -Suggested improvement

%Raw data reporting&Pressure volume (experimental data) &Differential liberation (experimental data)&Viscosity (experimental data)

%Data consistency&Mixture density calculation & verification&Co calculation & verification

ConclusionsConclusionsMore improvement in the following areas:

%% Problems related to experimentationProblems related to experimentation&Reservoir process presentation

&&Physical trends of lab dataPhysical trends of lab data

%% Problems related to calculationsProblems related to calculations&Adjustment of differential data

% Problems related to Problems related to data smoothingsmoothing&Y-function

&XY-function

ConclusionsConclusions%% Problems related to correlationsProblems related to correlations

&Physical trends of correlations

&&Pitfalls of least square methodPitfalls of least square method

% Artificial neural networks&Design of ANN

&Over Fitting

%% PVT ReportingPVT Reporting&Raw data reporting

&Data consistency

Final CommentFinal Comment

There are challenges in addressing these

problems, but there are untapped scientific

tools as well.

We explored these challenges and

examined possible solutions.

Thank YouThank You

!"#$