You are on page 1of 8

SPE

SPE 11148 The Accuracy of Pulsed Neutron Capture Logs for Residual Oil Saturation
by Philip A, Schenewerk and Roy M, Knapp, U, of Oklahoma, and Walter H. Fertl, Dresser At/as Petroleum Engineering Services Members SPE

This paper was presented al the 571h Annual Fall Techmcal Conference and Exhibition of the Society of Pelroleum Engineers of AIME, held in New Orleans, LA, Sept. 26-29.1982. The material is sub}ecl 10correction by the aulhor. permission 10copy IS restricted 10an Drawer 64706, Dallas, Texas 75206. abstract of not more than 300 words. Wrife: 6200 N. Cenfral Expressway, P.O.

ABSTRACT Pulsedneutroncapture(PNC) logshave been used to determine residual oil saturations for many yea:s. A previousstudyfoundthat at low values of residual oil saturation (ROS)conventional PNC loggingtechniques did not have the accuracy necdeci8ion making essaryfor enhancedoil recovery requirements. Speciallog-inject-loe techniques
were developed in order to reduce the uncertainty In valuee of ROS measured with PNC logs.

1. Reservoir prospect screening basedon productionand injection history,geology, reservoir, and fluidproperties, 2. Pre-pilot evaluation baaedon pressure teete and infillwells for specialcoringand loggingprocedures. 3. A fieldpilottest to determine
efficiency. 4. The commercial venture decision the results of steps 2 and 3. recovery

based

upon

A study

of the uncertainty

associated

with

ROS

valuesdetermined with PNC logswas made using MonteCarlosimulation techniques.Fielddata was obtained from testsreported in the literature. Usinga more realistic confidence interval of 95%, of the ROS valueswas found the expected accurace to decrease by at leasta factorof two in all tests. Additional uncertainty introduced by the log-injectlog process was not modeled. INTRODUCTION Enhanced oil recovery, or tertiary recovery, is en attemptto recoverthe oil remaining in the reservoir at the end of primaryand secondary recovery. This remaining oil saturation is called the residual oil saturation (ROS), Enhancedoil recovery techniques includethreeclasaesof processes:thermal, chemical, and miscible, The polymer, and chemical technique are surfactant, of micellar alkaline flooding. The injection chemicale, carbondioxide, and flue gas comprise the suiteof miscibletechniques.Theseprocesses have been described in the litaratura, 6 Not all are applicable to any one reservoir, so techniques screening criteria hava been developed to determine for a given::: of reathe appropriate technique ervoirrockand fluidcharacteristic, the staps involved in Hasibaet alg outlined porject. Thsy planning end enhanced oil recovery
are:

The ultimatedecision made in the finalstep depends on the two key parameters (1) ROS, the amount of oil left in placeat the end of primaryand secondary recovery, and (2) the recovery efficiency, the amountof the remaining oil wb.ch will be recovered. ROS can be dete:nined by several methods,the most promising of which la well logging. One of the most usefulloggingtoolsis the pulsedneutroncapture log (PNC)since it can be run in both open and casedwellborea.Well logging methodsare popularbecauaethey enablethe engtneerto observe and monitorverticalsaturation profilesin the well. When used in multiple wells it is then possible to determine lateral variationa in saturations, For a technique to be usedproperlyit ia important to know the associated limitations and the overallaccuracy conatrainta.This is especially sincea multi-million true of ROS determination dollardecision restson the value. The objeutive of this studyla to assesathe accuracy of both PNC logs and the specialtechniques developed to determine ROS, This assaesment will then allow the user to know what levelof confidence of these techniques can be placedin the resultant ROS values. PNC LOG DETERMINATION OF ROS Pulsedneutroncapture(PNC)logswere developec in the early 1960sto meaeure water saturations in high poroeityformation containing high ealinity wat.era.]z The toolmeeeuresthe totalor bulk cap-

Referencea end Illustrationat end of paper.

THE ACCURACY OF PULSEDNEUTRONCAPTURELOGS FOR RSSIDUAL OIL SATURATION nETERMTNATTONS -. _-

SPl?

lllfin

:urecross sectionof the formation in both open and :aeed well bores. The overallresponse in a shale Freereservoir can be expressed as: b= ~ma (1-$)+ ZWSW$+ ~hc (l-SW)$ (1)

5. The injection is radially complete and exceedsthe depthof investigation of the tool .
6.

The bottomhole injection pressureis below the formation fracture pressure.

~orthe conventional determination of ROS usingpulsed leutron capturelogsequation(1) ia rearranged yieldLng

ROS=l-S

=1-

b- ma+ $ Zma-he) $(ZW - Xhc)

(2)

Even usingthiswaterflood log-inject-log technique thereis stillsome uncertainty in the value of ROS, Robinsone did experimental work with a modified PNC tool in an effortto furtherreducethe uncertainty in ROS estimation.This modifiedtool was used to make stationary measurements. MONTECARLOSIMULATION

When precisevaluesare known for each parameter involvedin eitherof the equations presented for ROS determination, then a singleprecisevalueof ROS can be calculated.Unfortunately, uncertainty is associatedwith each of the required parameters.This uncertainty resultsfrom two sources. The firstis are measuredand the fact thatsome of the parameters thereis alwayasome uncertainty in the measurement procass. The secondaspectof the uncertainty is a The parameter leaetlikelyto be known ie the resultof the problemwhichariseswhen reservoir matrix capturecrosssectionsEm . This parameter parameter are not knownprecisely andmuet be eeti:anbe estimated from PNC log da?a in eitheran mated. Walstromet al presented a methodfor deter~djacent water zona of knownporosityor a densezone. miningthe value of a function when thereie uncer[n either case the assumption must be made that the tainty in the input paremetera. This method 18 called formations are made up of the samematrixmaterial. MonteCarlosimulation. tecently, techniques have becomeavailable for meas~ringmatrixcapturecrosssectionsfor smallcore In a Monte Carlosimulation, a mathematical model ~emples.g the processor operation ie developed whichdescribes of intereat. The model is then used to performa Youmanset al2proposeda processfor ROS numberof repeated experiments or trials. For each determination with PNC logs. The waterflood logtrialthe imputparameters are sampledfrom their Lnject-log technique reducesthe uncertainty in ROS respective probability distributions in some random >y eliminating the need for a knowledge of both the fashion. The experiment is performed and the trial hydrocarbon matrixcapturecroaesections. This resultaare analyzedusingstatistical techniques. approach involvesloggingthe formation which results l%is approachie usedregularly in the petroleum in industryto evaluatethe economic attractiveness of explor~tion prospects, workovers, and secondary or (3) tertiary recovery projects.1 s Monte Carlosimulabl= ma (1-$)+ZWISW$+ hcso$ tion techniques can alsobe used to studythe uncertaintyin ROS valuesdetermined fromPNC logs. In lhen a water of contracting salinityis injected and additionto determining a mean valueof ROS the the formation is relogged whichyielda techniqu? also yieldsa distribution function from which confidence interval information can be obtained (4) b2= ma (1-40+2W2SW4+ hcso$ Probability distributions are used to expressthe combining equations (3) and (4) and solvingfor ROS of interest. Although uncertainty in some parameter resultsin many distribution functions have baen proposed, this studywill uae uniformand triangular distributions b2 - bl (5) ROS = l-SW= 1 for variableuncertainty.The particular distribu$ (~w2-~wl) tionschosenshouldreflectthe accuracy with which the parameter ie knownor underetood,s l%is technique is baaedon severalassumptions.They are: The uniformprobability distribution la chosen when a parameter is confined betweensome upperand 10 No free gas is present. lowerlimit, Everyvaluaof the parameter between thoselimitshas an equallylikelyprobability of oil occurring.The cumulative probability of a parameter 2. Tha formation is at its residual saturation, X is givenbys 3. There ie no changein oil saturation due to the injection of fluid. 4. There is no shrinkage of tha reservoir oil. F(x) - ~ (6)

In orderto determine ROS usingequation(2). :Ivevariables must be known. Multiplepassesof the logging tool are made acrossthe formation in order :0 reducethe uncertainty in Xb, the bulk capture :roas section. The capturecrosssect$ons of the !ormation water,Zw, and hydrocarbons, Z , can be determined from fieldmeasurement if euf k$cient if not, the inluantities of fluidsare available; \ormation can be obtainedfrom laboratory analysis >rby usingpublished correlations. Porosity, $, :enbe obtainedfromeitherwell log or core data.

By replacing F(X) with a uniformly distributed random

wmber R, then solvingfor X the equation becomes X= Xt+ R(% - X2) (7)

by the personmaking the estimation.It shouldbe noted thatseveralauthorshave pointedout that there is a tendency to underestimate uncertainty whichmakes the parameters probability distribution too narrow, 2 The uncertainties in ROS reported in this study are based on confidence intervals determined from the frequency distribution generated by the Monte Carlo model, For each set of data from the fieldtests two model runswere made in orderto determine the overallrangeof uncertainty in ROS. These two runs were calledthe best and worst cases. The best case model run was made usin~the assumption thateach parameters uncertainty couldbe approximated by a triangular distribution.The worst casemodelwas run usingthe uniformdistribution to model the uncertainty in each parameter.When only the modeling parameters are considered, the actualuncertainty in ROS lies somewhere betweenthe best and worst case uncertainties. The totaluncertainty in ROS may not be due to parameter uncertainty alone. The interpretive equations previously developed have some simplifying assumptions incorporated into themwhichmay or may not be true depending on the particular fieldtest. When the aesumptione required for the interpretive equations are not true,the uncertainty may actually be higherthan the worst casemodel indicatee. A. Conventional Water Saturation Determination

The triangular probability distribution is used #hena parameter has an upperand lowerboundas well 1s a most likelyvalue. The cumulative probability >f X is givenby= X-xg z Xm-Xk (8) .- F(X) = -k %-xt

F(x)

.1-

2 \-xm l-xm 5-XE $-x

(9)

#henX <X<. By replacing F(X)with a uniformly iistri~u~ed~an om numberR and solvingfor X the ?quations become

?4

X= XL{(Xm-XL) (%-XL)

d
(l-R))%

(lo)

#henR: {(Xm-XQ) / (~-X1)} and X=~-{(~-Xm) (~-Xg)


(11)

#henR: {(Xm- Xg) / (~ - XL)}. MODELAPPLICATIONS AND RSSULTS Richardson et al used conventional PNC logging techniques to determine the ROS valuein a Miocene unconsolidated sand reeervoir. In the test the bulk capturecrosssectionfor the two zonesof interest was determined from ten repeatpassesover an adjacent contribuwater zone. Table 1 showseach parameters tion to the overalluncertainty in the resultant value of ROS.

In order to study the uncertainty inherent in ROE valuesdetermined by PNC logs,a Monte Carlo simulation model waa developed for both the conventionalwater saturation determination and the waterfloodlog-inject-log technique.Fielddatawas obtained from testsreportedin the literature. Richardson et at presented data for both the conventionalwater saturation determination and the Table 4 shows the resultsof the MonteCarlo waterflood log-inject-log technique usingPNC logs. modelingof this fieldtest. Modeledresultsin Thesedata are re ortedas Tables1 and 2, respecuncertainty at one standard deviation were similar ~ reported tively. Robinson data froma waterflood to previously published values;therfore, only one log-inject-log testusinga modifiedPNC loggingtool. value is reported. At a realistic levelof confidenc< The data from this test is shownin Table 3. (i.e.95%) the expectedaccuracy for ROS valuea decreases.In Zone A the expected accuracyfor ROS The valuesof each parameter and its associated decreases to iO.244at best and kO.317at the worst. uncertainty were determined in one of two ways, Zone B showsa decreasein the expected accuracy in eitherby measurement or by estimation.When a ROS to *0,255at beat and +0.366in the worat case. parameter was measuredin the field,ita valuewas determined by multiple measurements.For example, in B, Waterflood Log-Inject-Log the conventional PNC log application reported by Richardson et al the valueof the bulk capturecross Richardson et al usad the waterflood log-inject section was determined from ten repeatpaaaesof the log technique with PNC logs to measureROS in three loggingtoolover each zone. From thesemultiple fieldtests. The te9twas suc~essful in only one wel: measurements it is poaaibleto obtaina mean value in which threeintervala were examined, Multiple alongwith an associated standard deviation.The rspeatpasaesof tha loggingtool were used to deterlimitsof the parameters probable rangeof values mine the bulk capturecrosssectionprior to and afte; can be obtained by addingto and subtracting from water injaction,Table2 showaeach paremetera the mean a valuewhich is 3.09 timssthe standard in the recontribution to the overalluncertainty deviation.* sultantvaluesof ROS. may be estimatad when The value of a parameter it in not possible to measureit d%rectly.When a parametera value is estimated, it ia eitherbaaed on fieldexperience or it is estimated throughthe In eithercaae the corralationa. use of generalized rangeare also d~termined limitsof the parameters *TMS yields a 99,9%confidence interval, Table 5 showsthe modelingresultsof this test. Again,the ROS valueaand the uncertainty valuesat one standard daviation were the samewith the exceptionof Zone 3. Afterexamination of the original data it waa felt that the original ROS valuewas probablyin error.

THE ACCURACYOF PULSEDNEUTRONCAPTURELOGS FOR RESIDUAL OIL SATUIbiTION DETERMINATIONS SPE 11148 Sw - watersaturation, fractional pore volume \ - upper11 mit of a parameter distribution k - lowerlimitof a parameter distribution

Afterassessing the uncertainty at a realistic level,it can be seen that the uncertainty in ROS has increased by a factorof two for a 9577 confidence interval. The expected accuracy for ROS in Zone 1 will decrease to *0.163at best and iO.205at worst. In Zone 2 the expectedaccuracy for ROS will decrease at best and tO.251at worst. The expected to to.209 to tO.209at best and accuracy for ROS decreases iO.251at worat in Zone 3. used a modified PNC log to performa Robinson waterflood log-inject-log determination of ROS in swell in southern Louisiana,The contribution of eachparameter to the uncertainty in ROS is shownin rable3. It is obviousthat this technique has reducedthe uncertainty in the loggingmeasurements by a substantial amount.

Xm - most likelyvalueof a parameter distribution b - bulk macroscopic thermalneutroncapture crosssection, S.U.

z - hydrocarbon macroscopic thermal neutron hc capturecrosssection, S.U. z - matrixmacroscopic thermalneutroncapture ma crosssection, S.U.

z - watermacroscopic thermalneutroncapture The simulation modelresultsare shownin Table w crosssection, S.U. 6. Robinson* did not consider uncertainty in porosity values. When this additional uncertainty is consid$ - porosity, fractional bulk volume ered the overalluncertainty increases.In Zone 1, the expected accuracy has decreased to ?0.067at best REFERENCES and tO.094at worat for a 95% confidence interval. Therehas been a decrease in the expected accuracy 1. Anderson, FL L., Application of Risk Analysisto of ROS to iO.071at best and tO.099at worst in Zone EnhancedRecovery Pilot TestingDecisions, 2. Zone 3 showsa decrease to tO.08at best and J, Pet. Tech.(December, 1979):1525-1530. tO.11at worst in expected ROS accuracy. CONCLUSIONS 2. Capen,E. C., The Difficulty of Assessing Uncertainty, J. Pet. Tech.(August, 1976): 843-850. Haaiba,H. J.; Wilson,L. A.; end Martinelli, J. W., How to Organize and Plan Enhanced Recovery Efforts, World Oil (January,1977): 91-95. Hernkel,J., Projections of Enhanced Oil RecoveryL 1985-1995,DoE/EIA-o183/11, 1979. McCray,A. W., Petroleum Evaluations and Economic Decisions. Englewood Cliffs:PrenticeHall, 1975. Poettmann, F. H. cd., Secondary and Tertiary Oil Recovery Processes, Oklahoma City: Interstate Oil CompactCommission, 1974. Richardson, J. E,; Wyman,R. E.; Jordan,J, R.; and Mitchell, F. R., Methodsfor Determining Residual Oil With PulsedNeutronCapture Logs,J. 1973):593-606. . Pet, Tech.(May, Robinson, J.D,,Neutron DecayTime in the Subsurface: Theory,Experiment, and Application to Residual Oil Determination, SPE 5119* presentad at 49thFall SPE Meeting, Houston, 19740 SPWLA/DOE, NeutronPorosity Modeling Workshop, Tulsa,January27, 1982. Waletrom, J. l%.; Mueller, T, D.; and McFarlane, R. C., Evaluation Uncertainty in Engineering Calculations, J. Pet. Tech, (December, 1967):1595-1603.

PNC logs are a valuable tool in the estimation of ROS for enhanced oil recovery projects. The fieldteatastudiedin the reportwere all performed 3. usingthe PNC logs of the early 1970s. Since that time thesetoolshave been replaced with improved versiona which shouldimprovetheiroverallaccuracy. However, severalconclusions can be made. 4. 1. The chosenlevelof significance determine the uncertainty in the estimate of ROS measured with PNC logs. Previousfieldtests 5, reported uncertainty in ROS at confidence levelsof one standard deviation (~68%). At a more realistic confidence level (95%), b the uncertainty in ROS is at leasttwo times higherthanpreviously published.
G

2, Unlessaccurate valuesof matrixcapture crosssections are available, conventional 7. applications of PNC logswill probably not yieldROS valuesat acceptable levelsof uncertainty. 3, PNC log improvements can only reducethe uncertainty in ROS to a certainlevel. This 8, ia becausein all caaesporosity information is required which thenbecomesthe limiting factor, 4, This studyfocusedonly on parameter meaeurement and ita aaaociated uncertainty.The uncertainty in ROS determined by waterflood 9. log-inject-log proceeees is also affected by stripping, shrinkage and incomplete dis10. placement which all act to Increaee the overalluncertainty. NOMENCLATURE so - oil saturation, fractional pore volume

11, Youmane,A. H; Hopkinson, E. C.; Bergen,R. A.; md Oshry,H, L., NeutronLifetime a New

I 11148

P* NuclearLog,J. Pet. Tech. (March,1964): 319-328.

12* Youmans,A. H.; Hopkinson, E. C.; and Wichmann, Lifetime Loggingin Theory P. A., Neutron and Practice, presented at 7th Annual SPWLASymposium, Tulsa,1966.

Table 1 Conventional Water Saturation Determination Reservoir Parameters and Contributions to Uncertainty Best Estimate in Parameter 21.6 S.U. 11.9S.tl. 87.0 S.U. 20.5S.Il. 0.29 28,3 S,U, 11,9au. 87.0 S.U* 20,5 S.U. 0,29 Uncertainty in
Best Estimate

Interval Zone A

Parameter b E ma z w hc o

Uncertainty

Contribution to in ROS* *0*094 fO.316 *0.011 io.oo5 ?0.035 *O*122 *0.316 to*022 +0,002 ?0,059

il.79S.u. *8.19S.U. t2.oo S.u, tO.50au. *0.02 ?2.35S.U. *8.19S.11. *2,00 S.u, *()*50 Souo io*02

ZoneB

b z ma z w E hc 4

S.U, - sigma unit ~ fractional pore volume

Table 2 Waterflood Log-Inject-Log ROS Determination Reservoir Parameters and Contributions to Uncertainty

Best

Interval Zone 1

Parameter

Estimate in Parameter 17.822


S.U. S.U. SOU.

Uncertainty in Best Estimate W0593 12.871 tl.313 *3.075 fool S-U*


S*UO

Contribution to Uncertainty in ROS +0.042 +0.201 iO.006 *0*014 to. 033

b 1 E b2

29.535 42.500 99.500 0.25

zW1 xW2

S*U* S*U*

S.u.

Zone 2 b2

18.824 32.204 42.500 99.500 0.27

SOU. S.Cl. S.U.

*0.649 *4.178 *1.313 f3.075 *0.01

S.Il.

t0,042 *O. 272 +0.006 *0.015 +0.032

SOUO Souo S*U.

S*UO

Zone 3

b 1 b2 E W1

18.824 32.222 42.500 99,500 0.27

S.U. S.U. S.UO

fO.667 *4*175 *1,313 k3*075 *0.01

S.UO

*0.044 fO.271 fo,oo? *0*015 f0,032

S*U* S*U*

zW2

SOU*

Table 3
Log-Inject-Log Waterflood ROSDetermination

Reeervolr Parameters and Contributions


Best Estimate Interval Zone 1 Parameter b 1
b2
z

Using a Modified PNCLog to Uncertainty

in

Parameter
8,U. S,U.

Uncertainty in Best Estimate

Contribution
Uncertainty tOc C132

to

in ROS

18,070

*0.439 S,uo to,739 S,u.


S,U, 10.936

27.966
31,532 73.387 0.325

*0.054 +0.016 tO.038 *O. 045

W1

a,u.
S.U,

E W2 $

*2,172
io,

S,U,

02
S,U,

Zone

b 1 b2
z

16.908 26,639 31.532 73.387 t), 325

S.U. S,(I, S.U. S.U.

*0.615 ?0,717 *0.936 *2.172

to, 045 ?0,053 io.037 *0,037 *0.044

S,u, e,u,
S,U.

W1 W2

*0.02
S.U.

Zone

b 1
b2
z

16.588

tO. 538 au. ?O.825 e,u. *0.936 *2.172


*O. S.U. S.U.

*0.040 *0,061 io. f139


*0.046
50.055

28.580

S,U.

W1 E W2
@

31.532 au,
S,U. 73.387

0.325

02

Teble

ROSEstimatea Interval Zone A

Using Conventional ROS

Water Saturation

Method

Uncertainty
*O.11O

in ROS
*

Richardson et al Best Caae


90% 95% 99%

0,63

0.627 0,627 0,627

tO.214 iO.244 +0,296

Worst Case
90%

95% 99% Zone B

0,627 0.627 0,627

iO. 286 io. 317 ?0.356

Richardson et al Beat Case

0,28

+0.120

90% 95% 99% Vors t Caee 90% 95x 99%

0.280 0.280 0,2s0 0.284 0,284 0.284

to, 220 t0,255 io,313 *O. 299 ?0, 336 ?0, 386

* at one standard deviation

~e%

Table 5 Waterflood Interval Zone 1 Richardson et al


Bast Case 90% 9577 99!!
Worst Case

Log-Inject-Log
.ROS 0,18

Estimatea

of ROS

Uncertainty

in ROS

?0>080 * *O, 142 *O, 163


+0.

0,177
0.177

0,177

194

90% 95% 99%

0.178 0.178 0,178

10.187 ?0.205 f0,233

Zone 2
Richardson Best Case 9077 95% 99%

et al

0,14
0.136 0.136 0,136

*0.090

tO. 182 *O. 209 t0,249

Table 6 Worst Caee


90% 95% 99% 0.148 0.148 0,148 tO.230 ~O. 251 fo.281

Waterflood

Log-Inject-Log

Eatimetea

of ROSUsing a Ftodified PNC Log


ROS

Interval

L%certaintY in ROS *0.025 ?0.058 to.067 to*086 ~0.081 to.094 iO. 116 tO.026 *0.061 to,071 *0.091 t0,085 50.099 :0,122 +0,028 *O!068 t0,080 *O,1O1 *0,095 *O,11O *O. 137

Zone 3
Richardson Beat Caae 90% 95% 99X Worst Case 90% 95% 99x et al 0.15 *O.1OO *

Zone 1 Robinnona
0,274

0.135 0.135 0.135

*O. 182 *O. 209 *O. 249

Eaet Caea
90% 95% 99x Woret 0.270 0.270 0.270

0.147
0.147 0.147

tO.230 to.251 ~0.281

Case
0.271 0.271 0.271

90% 95x 99% Zone2 Robinson BeetCase 90% 95% 99% Worst Caee 90% 95% 99% Zone3 Robinson BestCaae 90% 95% 99% Worst Case 90% 95% 99X

* at one at~dard

deviation

%8%

0,286

0,284 0,284 0.284

0,283 0,283 0,283

0,120

0.117 00117 0, 117

0,11? 0.117 0,117

You might also like