Modeling of Steam Distillation Mechanism during Seam Injection Process Using Artificial Intelligence

Research ArticleModeling of Steam Distillation Mechanism during SteamInjection Process Using Artificial Intelligence

Amin Daryasafar Arash Ahadi and Riyaz Kharrat

Petroleum Department Petroleum University of Technology PO Box 6198144471 Ahwaz Iran

Correspondence should be addressed to Amin Daryasafar amindaryasafaryahoocom

Received 27 February 2014 Revised 15 March 2014 Accepted 29 March 2014 Published 28 April 2014

Academic Editor Nirupam Chakraborti

Copyright copy 2014 Amin Daryasafar et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

Steam distillation as one of the important mechanisms has a great role in oil recovery in thermal methods and so it is important tosimulate this process experimentally and theoretically In this work the simulation of steam distillation is performed on sixteen setsof crude oil data found in the literature Artificial intelligence (AI) tools such as artificial neural network (ANN) and also adaptiveneurofuzzy interference system (ANFIS) are used in this study as effective methods to simulate the distillate recoveries of thesesets of data Thirteen sets of data were used to train the models and three sets were used to test the models The developed modelsare highly compatible with respect to input oil properties and can predict the distillate yield with minimum entry For showingthe performance of the proposed models simulation of steam distillation is also done using modified Peng-Robinson equation ofstate Comparison between the calculated distillates by ANFIS and neural networkmodels and also equation of state-basedmethodindicates that the errors of the ANFIS model for training data and test data sets are lower than those of other methods

1 Introduction

One of the most successful methods for heavy oil productionis steam flooding and while steam has been injected intoreservoirs almost as long the mechanisms of this processare much less understood Several experimental studies areperformed for studying the effect of different mechanismssuch as viscosity reduction wettability alteration and steamdistillationvaporization during this method of recoveryAmong all these mechanisms steam distillation mechanismis the main difference between steam and other thermalmethods

Steam distillation process happened when light fractionsof crude oil are separated by injecting the steam into the crudeoil Observation of the produced vapors of matured steamfloods proves the fact that steam can carry a large amount oflight hydrocarbons in the steam distillation process Severalpapers have reported the effects of steam distillation on oilrecovery observed in laboratory steam displacement testsFarouq Ali [1] estimated that 5 to 10 of the heavy oilrecovery and as much as 60 of the light-oil recovery maybe attributed to steam distillation mechanism Willman et al

[2] demonstrated that steam flooding produces significantlygreater oil recovery than that in flooding with hot water at thesame temperature Mainly this is due to steam distillationWu and Fulton [3] reported that oil in the steam plateau ofan in situ combustion process is removed mainly by steamdistillation Johnson et al [4] showed that the oil vaporizationrecovery by steam ranges from 547 to 940 of immobile oilvolume

Several methods have been presented for simulatingsteam distillation mechanism in steam injection processSukkar [5] used the relative velocities of steam the steamfront and also the rates at which hydrocarbon componentswere distilled to estimate the amount of oil distilled duringsteam flooding Holland and Welch [6] developed a modelfor calculating steam distillation yield at saturated steam tem-peratures where the solubility of hydrocarbon and water isnegligible Duerksen andHsueh [7] proposed correlations forthe prediction of steam distillation yield with different crudeoil properties and operating conditions They also showedthat the distillation recovery correlates well with AmericanPetroleum Institute (API) gravity and wax content Northropand Venkatesan [8] presented an analytical multicomponent

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 246589 8 pageshttpdxdoiorg1011552014246589

2 The Scientific World Journal

Table 1 Trial and error calculations for selecting the most suitableANN

Number of neurons inthe hidden layer

Training data(RMSE)

Test data(RMSE)

5 00122 003167 00126 0039 00125 0030211 00121 0031415 00115 0031218 00118 0031120 00116 0029721 00119 0029823 00129 0031724 00141 0031925 00132 00312

model to predict steam distillation yield and showed thatthe distillation yield increases as the temperature increasesVan Winkle [9] proposed a method to predict the amountof steam required for distillation of a specific amount of avolatile material based on Raoultrsquos and Daltonrsquos laws

Thementioned studiesmay face considerable errorswhenthey are applied to crude oil samples and some of themrequire experimental data such as oil characterization dataso we need to propose a model for prediction of steamdistillation yield with minimum entry data

The complexity of steam distillation mechanism leadsus to use artificial intelligence such as artificial neural net-work (ANN) and adaptive neurofuzzy interference system(ANFIS) for simulation of steam distillation process Inthis paper we use ANN and also ANFIS to propose apractical model for predicting the steam distillation recoveryas accurate as possible by choosing the best model basedon laboratory data This model can be applied to predictthe steam distillation yield of crude oils with new properties(Table 2)

2 Description of Method

21 Artificial Neural Network (ANN) A neural network isstructured by multiple connection units arranged in layerswhich indicate the weights between neurons that are learnedunder an optimization criterion ANNs provide a nonlin-ear mapping between inputs and outputs by its intrinsicability [11] The success in obtaining a reliable and robustnetwork depends on the correct data preprocessing correctarchitecture selection and correct network training choicestrongly [12] Artificial neural networks have been developedfor a wide variety of problems such as classification func-tion approximation and prediction Multilayer feedforwardnetworks are the most commonly used for the functionapproximation Feedforward networks consist of groups ofinterconnected neurons arranged in layers correspondingto input hidden and output layers Once the input layerneurons are clamped to their values the evolving starts layerby layer and the neurons determine their output and this

is the reason that these networks are called feedforwardThe dependence of output values on input values is quitecomplex and includes all synaptic weights and thresholdsUsually this dependence does not have a meaningful analyticexpression These types of network can approximate mosttypes of nonlinear functions irrespective of how much theyare complex

The network is trained by performing optimization ofweights for each node interconnection and bias terms untilthe obtained values of output become as close as possible tothe actual outputs

The type of artificial neural network used in this studywasMultilayer feedforward networkWe need enough exper-imental data for training the network Sixteen experimentaldata sets were used for simulation of steam distillation in thisstudy and these data sets are obtained from literature [10]Thirteen crude oil data sets were used as training data and thedata sets related to Shiells Canyon Teapot Dome and RockCreek oil fields were considered as test data The inputs ofthis network are American Petroleum Institute (API) gravitykinematic viscosity at 378∘C characterization factor andsteamdistillation factor while the output is distillate recoverySteam distillation factor is the ratio of the volumetric amountof steam injected based on cold water equivalent and thevolume of initial oil Distillate recovery is the volumetricamount of hydrocarbon distilled over initial oil volumeThe volumes are calculated at standard conditions Thecharacterization factor and API are defined as

characterization factor =average boiling point

specific gravity

API = 1415

specific gravityminus 1315

(1)

Levenberg-Marquardt back propagation algorithm was usedfor training the network [13] and the number of neurons inhidden layers was chosen according to the minimum rootmean square error (RMSE) by trial and error

RMSE =

norm (simulated result minus experimental result)sqrt (length (simulated result))

(2)

Table 1 shows the results of trial and error calculationsused in this study for determining the number of neurons inthe hidden layer Several networks were trained and finally anetwork with one hidden layer with twenty tangent sigmoidneurons was selected as the most suitable network Theneurons in the output layer have linear transfer functionsThetangent sigmoid function is defined as follows

tansig (119899) = 2

1 + 119890

minus2119899

minus 1 (3)

22 AdaptiveNeurofuzzy Inference System (ANFIS) Adaptiveneurofuzzy inference system (ANFIS) is a kind of neuralnetwork that is based on fuzzy inference system [14] Sinceit integrates both neural networks and fuzzy logic principlesit has potential to capture the benefits of both in a singleframework Generally two objectives are followed using

The Scientific World Journal 3

Table2Cr

udeo

ilprop

ertie

sand

experim

entalresultsof

steam

distillationrecovery

ford

ifferento

ilfieldsa

Num

ber

Field

Experim

entalresultsof

steam

distillationyields

for119881119908

119881119900119894

Crud

eoilprop

ertie

s1

23

45

1015

20API

Kinematicviscosity

at378∘

CCh

aracteriz

ationfactor

1SouthBe

lridge

0031

004

6006

0069

0075

01

0119

013

124

040

8597

2Winklem

anDom

e0089

0111

0125

0136

0142

017

0182

0195

149

004

8896

3WhiteCa

stle

007

0095

011

0122

0137

0185

021

023

1600308

974

Ediso

n0092

012

014

0151

0164

019

0198

0209

161

00397

975

RedBa

nk0128

0162

018

0195

0205

0231

0241

025

171

003

996

Slocum

0032

008

0097

011

0122

0172

0195

02

188

00395

100

7HiddenDom

e0119

0148

0169

019

0205

025

028

0295

207

00086

101

8To

borg

0196

0239

0267

0285

03

0339

0349

036

222

00036

101

9Brea

021

024

0265

0283

0296

033

034

0354

235

00039

100

10Shanno

n014

0192

022

024

026

0307

0328

0331

247

00032

102

11Ro

binson

0128

0176

0208

0228

0245

0295

0312

032

2600029

103

12El

Dorado

0345

04

043

044

1045

047

0475

048

325

000

05101

13Sh

iells

Canyon

0378

0438

047

049

0508

0541

0558

057

33000

06102

14Teapot

Dom

e024

032

036

0396

0425

0503

0534

057

345

000

06104

15Ro

ckCr

eek

0295

036

04

0412

042

044

70465

048

382

000

05104

16Plum

Bush

028

0338

036

038

04

046

0489

053

399

000

06105

a WuandElder1983

[10]


ANFIS integrating the best features of fuzzy systems andneural networks and their applicability to synthesize ANFIScombines the fuzzy logic if-then rules and the accuracyand learning power of neural networks to make them ahybrid intelligent system ANFIS has the ability to solvenonlinear problems For specifying the relationship betweeninput and output to determine the optimized distribution ofmembership functions two learning methods are generallyused in ANFIS These learning methods are back propa-gation and hybrid The hybrid system is a combination ofpropagation and least squares method [15] In the backwardpass the error is sent back through the network in a similarmanner to back propagation [16] Hybrid systems have beenused by researchers for modeling and predictions in variousengineering systems When generating a FIS using ANFISselecting proper parameters is very important including thenumber of MF for each individual antecedent variable andalso selecting proper parameters for the learning and refiningprocess Parameter selection and their impact on the ANFIShave been addressed in the literature [17ndash19]

For simulating steamdistillation process anothermodel isproposed usingANFIS For this purpose a structure with fourinputs with three 120587-shaped built-in membership functionswas considered FIS generation was done by grid partition-ing Grid partition divides the data space into rectangularsubspaces using axis-paralleled partition Π-shaped built-inmembership function is given by

119891 (119910119898 119899 119900 119901) =

0 119910 le 119898

2(

119910 minus 119898

119899 minus 119898

)

2

119898 le 119910 le

119898 + 119899

2

1 minus 2(

119910 minus 119899

119899 minus 119898

)

2

119898 + 119899

2

le 119910 le 119899

1 119899 le 119910 le 119900

1 minus 2(

119910 minus 119900

119901 minus 119900

)

2

119900 le 119910 le

119900 + 119901

2

2(

119910 minus 119901

119901 minus 119900

)

2

119900 + 119901

2

le 119910 le 119901

0 119910 ge 119901

(4)

The parameters 119898 and 119901 locate the ldquofeetrdquo of the curve while119899 and 119900 locate its ldquoshouldersrdquo We utilized a hybrid method[20] which is a combination of gradient method and leastsquares estimate (LSE) for training the system The inputs ofthe system are American Petroleum Institute (API) gravitykinematic viscosity at 378∘C characterization factor andsteamdistillation factor while the output is distillate recoveryAgain thirteen crude oil data sets were used as training dataand data sets related to Shiells Canyon Teapot Dome andRock Creek oil fields were considered as test data Schematicof the proposed ANFIS structure is shown in Figure 1

23 Equation of States Method The first step in simulationof the steam distillation process by EOS method is toevaluate the oil characterization This task is performed bydetermining data such as characterization factor averagemolecular weight viscosity API and distillation test dataFor determining the distribution of components in liquid and

vapor phases flash calculationmust be performed It must benoticed that several equations of states must be performedand then the best EOS will be chosen as the optimumequation for simulation of this process For this purposeseveral equations of states were tested in EOS method andaccording to the results the modified Peng-Robinson [21]equation of state seems to generate better results [22]

In this paper for better comparison between the proposedmodels and othermethods we usedmodifiedPeng-Robinsonequation of state to simulate the steamdistillationmechanismin steam flooding process For this purpose the multistageadiabatic flash calculation was performed In this processoil comes into contact with fresh steam in each stage andas equilibrium condition is reached the vapor phase whichincludes light fractions of oil and steam leaves the stage andthe remaining oil enters the next stage

The equation of state for mixtures proposed by Peng andRobinson [23] is as follow

119875 =

119877119879

119881 minus 119861119898

minus

119860119898

119881 (119881 + 119861119898) + 119861119898(119881 minus 119861

119898)

119861119898=

119873

sum

119894=1

119909119894119887119894

119860119898=

119873

sum

119894=1

119873

sum

119895=1

119909119894119909119895119886

0

119894119895

(1 minus 119896119894119895)

119886

0

119894119895

=radic120572119894120572119895119886119894119886119895

(5)

Mathias and Copeman [21] developed a density-dependentlocal composition (DDLC) model for the Peng-Robinsonequation of state Since themodel was too expensive for com-puter calculation they formulated the following truncatedmodel

119860119898=

119873

sum

119894=1

119873

sum

119895=1

119909119894119909119895119886

0

119894119895

(1 minus 119896119894119895)

minus

1

2119861119898119877119879

radic2

ln[119885 + 119861 (1 minus

radic2)

119885 + 119861 (1 +radic2)

]

timessum

119894

119909119894119886

2

119894

[

[

sum

119894

119909119895119889

2

119894119895

minus (sum

119895

119909119895119889119894119895)

2

]

]

119896119894119895= 119896119895119894 119889119894119895

= 119889119895119894

(6)

3 Results and Discussion

The system was trained several times to achieve the bestcorrelation between the simulated data and experimentaldata according to the value of mean square error (MSE) bothfor artificial neural network and ANFIS

In Figure 2 the best linear fit between the simulated andexperimental data is illustrated with correlation factor of09942 which indicates a very good correlation These resultsare obtained using ANNmethod

After training the network using ANN method thenetwork was performed on the test data and the simulationresults versus experimental test data are shown in Figure 3


Input Inputmf Rule OutputOutputmf

Logical operationsAndOrNot

Figure 1 The structure of ANFIS model for steam distillationrecovery estimation

01 02 03 04 05

005

01

015

02

025

03

035

04

045

05

Target = RF(exp)

Y=

RF(s

imul

ated

)

Ysim = 099lowast target + 00022

R = 09942

Data

Figure 2 The relation between ANN predictions and actualexperimental data

An ANFIS model was designed with four inputs (Amer-ican Petroleum Institute (API) gravity viscosity at 378∘Ccharacterization factor and steam distillation factor) eachwith three 120587-shaped built-in membership functions andone output (distillate yield) Figure 4 We utilized a hybridmethod for training the system The results of the trainingdata for simulated and experimental data are shown inFigure 5 which illustrates a very good correlation

For validation of the proposed model by ANFIS aftertraining the system it was performed on the test data and itsresult is illustrated in Figure 6

In this study we also used modified Peng-Robinsonequation of state which gives the best results than those of

0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very

ExperimentalSimulated

Figure 3 Performance of ANN for test data

others to simulate the steam distillation process Vafaei [22]found that the modified Peng-Robinson seems to generatebetter results and used this method to estimate the distillateyield For validation of their estimation and then for bettercomparison between the results obtained by different meth-ods and models we again used this kind of EOS to calculatethe distillate recovery The results are given in Table 3

The performance index used for evaluating the models isbased on the present of average relative deviation (ARE) as

ARE =

sum

119899

1

10038161003816100381610038161003816

(119881119900119881119900119894)sim minus (119881

119900119881119900119894)exp

10038161003816100381610038161003816

(119881119900119881119900119894)exp

119899

(7)

Table 4 shows the comparison between the results obtainedby different methods which were considered in this paperaccording to the obtained average relative error for bothtraining and test data

We must conclude that Vafaei [24] proposed a multilayerperceptronmodel for simulation of steam distillation processand used these sets of data for modeling this process buthe chose White Castle Toborg and Teapot Dome oil fieldsdata as test data and the remaining sets of data as trainingdata and their model obtained ARE of 747 and 1119 fortraining data and test data respectively but in this study wecould achieve the less AREby changing the conditions of eachsystem and also choosing different sets of data for trainingand testing

Comparison of the results shown in Table 4 proves thatthe proposed model by ANFIS gives better results for bothtraining and test data and also using artificial intelligence cangive better results with minimum entry without needing oilcharacterization

4 Conclusion

In this paper we proposed a model which can predict thedistillate yield in distillation process accurately using ANFISand ANN that are two important subbranches of artificialintelligence (AI) tools ANN is one of the effective tools forfunction approximation but it has some problems that ANFIScan solve for example for reaching the best network we


15 20 25 30 35

0

02

04

06

08

1

Input1

Deg

ree o

f mem

bers

hip

in1mf1 in1mf2 in1mf3

(a)

01 02 03 04

0

02

04

06

08

1

Input2

Deg

ree o

f mem

bers

hip


(b)

96 98 10 102 104

0

02

04

06

08

1

Input3

Deg

ree o

f mem

bers

hip


(c)

5 10 15 20

0

02

04

06

08

1

Input4

Deg

ree o

f mem

bers

hip


(d)

Figure 4 The generalized 120587-shaped membership functions of four input variables

0 20 40 60 80 100 1200

01

02

03

04

05

06

07

Data index

Reco

very

OriginalFIS predicted

Figure 5 Performance of ANFIS for training data

should run the systemmore times and in this studywe trainedthe network many times to reach the best model and this is atime consuming process but ANFIS removes this problem byfuzzification process

In this study we utilized a FIS structure with four inputseach with three 120587-shaped built-in membership functionsand hybrid method for training the system Thirteen sets ofdata were used as training data and three sets of data as testdataThe input data are the steam distillation factor viscosityAPI and characterization factor of the oil The obtainedresults by this method were compared with a multilayer feed

Table 3 Average relative error between simulated results by EOSand experimental data

Field ARESouth Belridge 1977Winkleman Dome 1987White Castle 3084Edison 1429Red Bank 1138Slocum 926Hidden Dome 281Toborg 858Brea 99Shannon 1167Robinson 2803El Dorado 4256Shiells Canyon 1376Teapot Dome 924Rock Creek 4589Plum Bush 3346

forward neural network and also an EOS-based methodThe comparison between the designed ANFIS and othertwo methods that is ANN method and EOS-based methodindicates that the accuracy of the proposed ANFIS modelfor both training and test data is better than that of othermethods Also both artificial intelligence models give betterresults than the proposed MLP model by Vafaei et al [24]


Table 4 Average relative error () between simulated resultsobtained by different methods and experimental data

Method Training data Test dataANFIS 201 612ANN 463 627EOS 1862 102

0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very

SimulatedExperimental

Figure 6 Performance of ANFIS for test data

Nomenclature

119886 Pure component energy parameter119886119894 Component 119894 energy parameter

119886119894119895 Composition and temperature dependent

binary energy parameter119886

0

119894119895

Binary temperature dependent energyparameter

119860119898 Mixture temperature dependent term

119887 Pure component volume parameter119887119894 Volume parameter for component 119894

119861 Dimensionless mixture parameter119861119898 Mixture volume parameter

119889119894119895 Binary interaction parameter forcomponents 119894 and 119895

119896119894119895 Binary interaction parameter for

components 119894 and 119895119875 Pressure119877 Gas constant119879 Temperature119881 Mixture molar volume119909119894 Mole fraction of component 119894

119885 Compressibility factor120572 Pure component temperature dependent

term120572119894 Temperature dependent term for

component 119894

Conflict of Interests

The authors declare that there is no conflict of interests regar-ding the publication of this paper

References

[1] S M Farouq Ali Jr Practical Considerations in Steam FloodingProducers Monthly 1968

[2] B TWillmanVVValleroy GWRunberg A J Cornelius andL W Powers ldquoLaboratory studies of oil recovery by steam inje-ctionrdquo Journal of Petroleum Technology vol 13 no 7 pp 681ndash690 1961

[3] C H Wu and P F Fulton ldquoExperimental simulation of thezones preceding the combustion front of an in- situ combustionprocessrdquo Society of Petroleum Engineers Journal vol 11 no 1 pp38ndash46 1971

[4] F S Johnson EGWalker andA F Bayazeed ldquoOil vaporizationduring steamfloodingrdquo JPT Journal of Petroleum Technologyvol 23 pp 731ndash742 1971

[5] J K Sukkar ldquoCalculation of oil distilled during steam floodingof light cruderdquo in Proceedings of the SPE Annual Technical Con-ference and Exhibition SPE 1609 pp 2ndash5 Dallas RichardsonTex USA 1966

[6] C D Holland and N E Welch ldquoSteam batch distillation calcu-lationrdquo Petroleum Refiner vol 36 pp 251ndash253 1957

[7] J H Duerksen and L Hsueh ldquoSteam distillation of crude oilsrdquoSociety of Petroleum Engineers Journal vol 23 no 2 pp 265ndash271 1983

[8] P S Northrop and V N Venkatesan ldquoAnalytical steam distilla-tion model for thermal enhanced oil recovery processesrdquo Indu-strial and Engineering Chemistry Research vol 32 no 9 pp2039ndash2046 1993

[9] M VanWinkleDistillation McGraw-Hill NewYork NY USA1967

[10] C H Wu and R B Elder ldquoCorrelation of crude oil steam dis-tillation yields with basic crude oil propertiesrdquo Society of Pet-roleum Engineers Journal vol 23 no 6 pp 937ndash945 1983

[11] K Hornik M Stinchcombe and H White ldquoUniversal approx-imation of an unknown mapping and its derivatives usingmultilayer feedforward networksrdquo Neural Networks vol 3 no5 pp 551ndash560 1990

[12] N Garcıa-Pedrajas C Hervas-Martınez and J Munoz-PerezldquoCOVNET a cooperative coevolutionary model for evolvingartificial neural networksrdquo IEEE Transactions on Neural Net-works vol 14 no 3 pp 575ndash596 2003

[13] K Levenberg ldquoAmethod for the solution of certain problems inleast squaresrdquoQuarterly of AppliedMathematics vol 2 pp 164ndash168 1944

[14] T Takagi and M Sugeno ldquoFuzzy identification of systems andits applications to modeling and controlrdquo IEEE Transactions onSystems Man and Cybernetics vol 15 no 1 pp 116ndash132 1985

[15] E D Ubeyli ldquoAdaptive neuro-fuzzy inference system employ-ing wavelet coefficients for detection of ophthalmic arterial dis-ordersrdquo Expert Systems with Applications vol 34 no 3 pp2201ndash2209 2008

[16] F Rahimi-Ajdadi andYAbbaspour-Gilandeh ldquoArtificialNeuralNetwork and stepwise multiple range regression methods forprediction of tractor fuel consumptionrdquo Measurement Journalof the International Measurement Confederation vol 44 no 10pp 2104ndash2111 2011

[17] B Fritzke ldquoIncremental neuro-fuzzy systemsrdquo in Proceedings ofthe Application of Soft Computing SPIE International Sympo-sium on Optical Science Engineering and Instrumentation pp86ndash97 San Diego Calif USA 1997


[18] J R Jang C Sun and E Mizutani Neuro-Fuzzy and Soft Com-puting A Computational Approach to Learning and MachineIntelligence Prentice Hall Englewood Cliffs NJ USA 1997

[19] M M Reda Taha A Noureldin and N El-Sheimy ldquoImprovingINSGPS positioning accuracy during GPS outages using fuzzylogicrdquo in Proceedings of the 16th International Technical Meetingof the Satellite Division of The Institute of Navigation (IONGPSGNSS rsquo03) pp 499ndash508 Oregon 2003

[20] J S R Jang ldquoFuzzymodeling using generalized neural networksand kalman filter algorithmrdquo in Proceedings of 9th NationalConference on Artificial Intelligence (AAAI rsquo91) vol 4 pp 762ndash767 1991

[21] PMMathias and TW Copeman ldquoExtension of the Peng-Rob-inson equation of state to complex mixtures evaluation of thevarious forms of the local composition conceptrdquo Fluid PhaseEquilibria vol 13 pp 91ndash108 1983

[22] M T Vafaei Simulation of distillation effects on oil recovery fac-tor during SAGD process [MS thesis] Shiraz University ShirazIran 2007

[23] D Y Peng andD B Robinson ldquoA new two-constant equation ostaterdquo Industrial ampEngineering Chemistry Fundamentals vol 15pp 58ndash64 1976

[24] M T Vafaei R Eslamloueyan and S Ayatollahi ldquoSimulation ofsteam distillation process using neural networksrdquo ChemicalEngineering Research and Design vol 87 no 8 pp 997ndash10022009

Submit your manuscripts athttpwwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014


Shock and Vibration


Mechanical Engineering

Advances in


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of


Distributed Sensor Networks


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Advances inOptoElectronics


Volume 2014

RoboticsJournal of


Chemical EngineeringInternational Journal of


Control Scienceand Engineering

Journal of


Antennas andPropagation




Navigation and Observation



Table 1 Trial and error calculations for selecting the most suitableANN

Number of neurons inthe hidden layer

Training data(RMSE)

Test data(RMSE)

5 00122 003167 00126 0039 00125 0030211 00121 0031415 00115 0031218 00118 0031120 00116 0029721 00119 0029823 00129 0031724 00141 0031925 00132 00312

model to predict steam distillation yield and showed thatthe distillation yield increases as the temperature increasesVan Winkle [9] proposed a method to predict the amountof steam required for distillation of a specific amount of avolatile material based on Raoultrsquos and Daltonrsquos laws

Thementioned studiesmay face considerable errorswhenthey are applied to crude oil samples and some of themrequire experimental data such as oil characterization dataso we need to propose a model for prediction of steamdistillation yield with minimum entry data

The complexity of steam distillation mechanism leadsus to use artificial intelligence such as artificial neural net-work (ANN) and adaptive neurofuzzy interference system(ANFIS) for simulation of steam distillation process Inthis paper we use ANN and also ANFIS to propose apractical model for predicting the steam distillation recoveryas accurate as possible by choosing the best model basedon laboratory data This model can be applied to predictthe steam distillation yield of crude oils with new properties(Table 2)

2 Description of Method

21 Artificial Neural Network (ANN) A neural network isstructured by multiple connection units arranged in layerswhich indicate the weights between neurons that are learnedunder an optimization criterion ANNs provide a nonlin-ear mapping between inputs and outputs by its intrinsicability [11] The success in obtaining a reliable and robustnetwork depends on the correct data preprocessing correctarchitecture selection and correct network training choicestrongly [12] Artificial neural networks have been developedfor a wide variety of problems such as classification func-tion approximation and prediction Multilayer feedforwardnetworks are the most commonly used for the functionapproximation Feedforward networks consist of groups ofinterconnected neurons arranged in layers correspondingto input hidden and output layers Once the input layerneurons are clamped to their values the evolving starts layerby layer and the neurons determine their output and this

is the reason that these networks are called feedforwardThe dependence of output values on input values is quitecomplex and includes all synaptic weights and thresholdsUsually this dependence does not have a meaningful analyticexpression These types of network can approximate mosttypes of nonlinear functions irrespective of how much theyare complex

The network is trained by performing optimization ofweights for each node interconnection and bias terms untilthe obtained values of output become as close as possible tothe actual outputs

The type of artificial neural network used in this studywasMultilayer feedforward networkWe need enough exper-imental data for training the network Sixteen experimentaldata sets were used for simulation of steam distillation in thisstudy and these data sets are obtained from literature [10]Thirteen crude oil data sets were used as training data and thedata sets related to Shiells Canyon Teapot Dome and RockCreek oil fields were considered as test data The inputs ofthis network are American Petroleum Institute (API) gravitykinematic viscosity at 378∘C characterization factor andsteamdistillation factor while the output is distillate recoverySteam distillation factor is the ratio of the volumetric amountof steam injected based on cold water equivalent and thevolume of initial oil Distillate recovery is the volumetricamount of hydrocarbon distilled over initial oil volumeThe volumes are calculated at standard conditions Thecharacterization factor and API are defined as

characterization factor =average boiling point

specific gravity

API = 1415

specific gravityminus 1315

(1)

Levenberg-Marquardt back propagation algorithm was usedfor training the network [13] and the number of neurons inhidden layers was chosen according to the minimum rootmean square error (RMSE) by trial and error

RMSE =

norm (simulated result minus experimental result)sqrt (length (simulated result))

(2)

Table 1 shows the results of trial and error calculationsused in this study for determining the number of neurons inthe hidden layer Several networks were trained and finally anetwork with one hidden layer with twenty tangent sigmoidneurons was selected as the most suitable network Theneurons in the output layer have linear transfer functionsThetangent sigmoid function is defined as follows

tansig (119899) = 2

1 + 119890

minus2119899

minus 1 (3)

22 AdaptiveNeurofuzzy Inference System (ANFIS) Adaptiveneurofuzzy inference system (ANFIS) is a kind of neuralnetwork that is based on fuzzy inference system [14] Sinceit integrates both neural networks and fuzzy logic principlesit has potential to capture the benefits of both in a singleframework Generally two objectives are followed using


Table2Cr

udeo

ilprop

ertie

sand

experim

entalresultsof

steam


ford

ifferento

ilfieldsa

Num

ber

Field

Experim

entalresultsof

steam

distillationyields

for119881119908

119881119900119894

Crud

eoilprop

ertie

s1

23

45

1015

20API

Kinematicviscosity

at378∘

CCh

aracteriz

ationfactor

1SouthBe

lridge

0031

004

6006

0069

0075

01

0119

013

124

040

8597

2Winklem

anDom

e0089

0111

0125

0136

0142

017

0182

0195

149

004

8896

3WhiteCa

stle

007

0095

011

0122

0137

0185

021

023

1600308

974

Ediso

n0092

012

014

0151

0164

019

0198

0209

161

00397

975

RedBa

nk0128

0162

018

0195

0205

0231

0241

025

171

003

996

Slocum

0032

008

0097

011

0122

0172

0195

02

188

00395

100

7HiddenDom

e0119

0148

0169

019

0205

025

028

0295

207

00086

101

8To

borg

0196

0239

0267

0285

03

0339

0349

036

222

00036

101

9Brea

021

024

0265

0283

0296

033

034

0354

235

00039

100

10Shanno

n014

0192

022

024

026

0307

0328

0331

247

00032

102

11Ro

binson

0128

0176

0208

0228

0245

0295

0312

032

2600029

103

12El

Dorado

0345

04

043

044

1045

047

0475

048

325

000

05101

13Sh

iells

Canyon

0378

0438

047

049

0508

0541

0558

057

33000

06102

14Teapot

Dom

e024

032

036

0396

0425

0503

0534

057

345

000

06104

15Ro

ckCr

eek

0295

036

04

0412

042

044

70465

048

382

000

05104

16Plum

Bush

028

0338

036

038

04

046

0489

053

399

000

06105

a WuandElder1983

[10]




119891 (119910119898 119899 119900 119901) =

0 119910 le 119898

2(

119910 minus 119898

119899 minus 119898

)

2

119898 le 119910 le

119898 + 119899

2

1 minus 2(

119910 minus 119899

119899 minus 119898

)

2

119898 + 119899

2

le 119910 le 119899

1 119899 le 119910 le 119900

1 minus 2(

119910 minus 119900

119901 minus 119900

)

2

119900 le 119910 le

119900 + 119901

2

2(

119910 minus 119901

119901 minus 119900

)

2

119900 + 119901

2

le 119910 le 119901

0 119910 ge 119901

(4)






119875 =

119877119879

119881 minus 119861119898

minus

119860119898

119881 (119881 + 119861119898) + 119861119898(119881 minus 119861

119898)

119861119898=

119873

sum

119894=1

119909119894119887119894

119860119898=

119873

sum

119894=1

119873

sum

119895=1

119909119894119909119895119886

0

119894119895

(1 minus 119896119894119895)

119886

0

119894119895

=radic120572119894120572119895119886119894119886119895

(5)


119860119898=

119873

sum

119894=1

119873

sum

119895=1

119909119894119909119895119886

0

119894119895

(1 minus 119896119894119895)

minus

1

2119861119898119877119879

radic2

ln[119885 + 119861 (1 minus

radic2)

119885 + 119861 (1 +radic2)

]

timessum

119894

119909119894119886

2

119894

[

[

sum

119894

119909119895119889

2

119894119895

minus (sum

119895

119909119895119889119894119895)

2

]

]

119896119894119895= 119896119895119894 119889119894119895

= 119889119895119894

(6)









01 02 03 04 05

005

01

015

02

025

03

035

04

045

05

Target = RF(exp)

Y=

RF(s

imul

ated

)


R = 09942

Data





0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very





ARE =

sum

119899

1

10038161003816100381610038161003816

(119881119900119881119900119894)sim minus (119881

119900119881119900119894)exp

10038161003816100381610038161003816

(119881119900119881119900119894)exp

119899

(7)




4 Conclusion



15 20 25 30 35

0

02

04

06

08

1

Input1

Deg

ree o

f mem

bers

hip


(a)

01 02 03 04

0

02

04

06

08

1

Input2

Deg

ree o

f mem

bers

hip


(b)

96 98 10 102 104

0

02

04

06

08

1

Input3

Deg

ree o

f mem

bers

hip


(c)

5 10 15 20

0

02

04

06

08

1

Input4

Deg

ree o

f mem

bers

hip


(d)


0 20 40 60 80 100 1200

01

02

03

04

05

06

07

Data index

Reco

very











0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very



Nomenclature




0

119894119895










component 119894



References



























VLSI Design



RotatingMachinery





Shock and Vibration



Advances in







Journal of





SensorsJournal of







Volume 2014

RoboticsJournal of





Journal of









Table2Cr

udeo

ilprop

ertie

sand

experim

entalresultsof

steam


ford

ifferento

ilfieldsa

Num

ber

Field

Experim

entalresultsof

steam

distillationyields

for119881119908

119881119900119894

Crud

eoilprop

ertie

s1

23

45

1015

20API

Kinematicviscosity

at378∘

CCh

aracteriz

ationfactor

1SouthBe

lridge

0031

004

6006

0069

0075

01

0119

013

124

040

8597

2Winklem

anDom

e0089

0111

0125

0136

0142

017

0182

0195

149

004

8896

3WhiteCa

stle

007

0095

011

0122

0137

0185

021

023

1600308

974

Ediso

n0092

012

014

0151

0164

019

0198

0209

161

00397

975

RedBa

nk0128

0162

018

0195

0205

0231

0241

025

171

003

996

Slocum

0032

008

0097

011

0122

0172

0195

02

188

00395

100

7HiddenDom

e0119

0148

0169

019

0205

025

028

0295

207

00086

101

8To

borg

0196

0239

0267

0285

03

0339

0349

036

222

00036

101

9Brea

021

024

0265

0283

0296

033

034

0354

235

00039

100

10Shanno

n014

0192

022

024

026

0307

0328

0331

247

00032

102

11Ro

binson

0128

0176

0208

0228

0245

0295

0312

032

2600029

103

12El

Dorado

0345

04

043

044

1045

047

0475

048

325

000

05101

13Sh

iells

Canyon

0378

0438

047

049

0508

0541

0558

057

33000

06102

14Teapot

Dom

e024

032

036

0396

0425

0503

0534

057

345

000

06104

15Ro

ckCr

eek

0295

036

04

0412

042

044

70465

048

382

000

05104

16Plum

Bush

028

0338

036

038

04

046

0489

053

399

000

06105

a WuandElder1983

[10]




119891 (119910119898 119899 119900 119901) =

0 119910 le 119898

2(

119910 minus 119898

119899 minus 119898

)

2

119898 le 119910 le

119898 + 119899

2

1 minus 2(

119910 minus 119899

119899 minus 119898

)

2

119898 + 119899

2

le 119910 le 119899

1 119899 le 119910 le 119900

1 minus 2(

119910 minus 119900

119901 minus 119900

)

2

119900 le 119910 le

119900 + 119901

2

2(

119910 minus 119901

119901 minus 119900

)

2

119900 + 119901

2

le 119910 le 119901

0 119910 ge 119901

(4)






119875 =

119877119879

119881 minus 119861119898

minus

119860119898

119881 (119881 + 119861119898) + 119861119898(119881 minus 119861

119898)

119861119898=

119873

sum

119894=1

119909119894119887119894

119860119898=

119873

sum

119894=1

119873

sum

119895=1

119909119894119909119895119886

0

119894119895

(1 minus 119896119894119895)

119886

0

119894119895

=radic120572119894120572119895119886119894119886119895

(5)


119860119898=

119873

sum

119894=1

119873

sum

119895=1

119909119894119909119895119886

0

119894119895

(1 minus 119896119894119895)

minus

1

2119861119898119877119879

radic2

ln[119885 + 119861 (1 minus

radic2)

119885 + 119861 (1 +radic2)

]

timessum

119894

119909119894119886

2

119894

[

[

sum

119894

119909119895119889

2

119894119895

minus (sum

119895

119909119895119889119894119895)

2

]

]

119896119894119895= 119896119895119894 119889119894119895

= 119889119895119894

(6)









01 02 03 04 05

005

01

015

02

025

03

035

04

045

05

Target = RF(exp)

Y=

RF(s

imul

ated

)


R = 09942

Data





0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very





ARE =

sum

119899

1

10038161003816100381610038161003816

(119881119900119881119900119894)sim minus (119881

119900119881119900119894)exp

10038161003816100381610038161003816

(119881119900119881119900119894)exp

119899

(7)




4 Conclusion



15 20 25 30 35

0

02

04

06

08

1

Input1

Deg

ree o

f mem

bers

hip


(a)

01 02 03 04

0

02

04

06

08

1

Input2

Deg

ree o

f mem

bers

hip


(b)

96 98 10 102 104

0

02

04

06

08

1

Input3

Deg

ree o

f mem

bers

hip


(c)

5 10 15 20

0

02

04

06

08

1

Input4

Deg

ree o

f mem

bers

hip


(d)


0 20 40 60 80 100 1200

01

02

03

04

05

06

07

Data index

Reco

very











0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very



Nomenclature




0

119894119895










component 119894



References



























VLSI Design



RotatingMachinery





Shock and Vibration



Advances in







Journal of





SensorsJournal of







Volume 2014

RoboticsJournal of





Journal of











119891 (119910119898 119899 119900 119901) =

0 119910 le 119898

2(

119910 minus 119898

119899 minus 119898

)

2

119898 le 119910 le

119898 + 119899

2

1 minus 2(

119910 minus 119899

119899 minus 119898

)

2

119898 + 119899

2

le 119910 le 119899

1 119899 le 119910 le 119900

1 minus 2(

119910 minus 119900

119901 minus 119900

)

2

119900 le 119910 le

119900 + 119901

2

2(

119910 minus 119901

119901 minus 119900

)

2

119900 + 119901

2

le 119910 le 119901

0 119910 ge 119901

(4)






119875 =

119877119879

119881 minus 119861119898

minus

119860119898

119881 (119881 + 119861119898) + 119861119898(119881 minus 119861

119898)

119861119898=

119873

sum

119894=1

119909119894119887119894

119860119898=

119873

sum

119894=1

119873

sum

119895=1

119909119894119909119895119886

0

119894119895

(1 minus 119896119894119895)

119886

0

119894119895

=radic120572119894120572119895119886119894119886119895

(5)


119860119898=

119873

sum

119894=1

119873

sum

119895=1

119909119894119909119895119886

0

119894119895

(1 minus 119896119894119895)

minus

1

2119861119898119877119879

radic2

ln[119885 + 119861 (1 minus

radic2)

119885 + 119861 (1 +radic2)

]

timessum

119894

119909119894119886

2

119894

[

[

sum

119894

119909119895119889

2

119894119895

minus (sum

119895

119909119895119889119894119895)

2

]

]

119896119894119895= 119896119895119894 119889119894119895

= 119889119895119894

(6)









01 02 03 04 05

005

01

015

02

025

03

035

04

045

05

Target = RF(exp)

Y=

RF(s

imul

ated

)


R = 09942

Data





0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very





ARE =

sum

119899

1

10038161003816100381610038161003816

(119881119900119881119900119894)sim minus (119881

119900119881119900119894)exp

10038161003816100381610038161003816

(119881119900119881119900119894)exp

119899

(7)




4 Conclusion



15 20 25 30 35

0

02

04

06

08

1

Input1

Deg

ree o

f mem

bers

hip


(a)

01 02 03 04

0

02

04

06

08

1

Input2

Deg

ree o

f mem

bers

hip


(b)

96 98 10 102 104

0

02

04

06

08

1

Input3

Deg

ree o

f mem

bers

hip


(c)

5 10 15 20

0

02

04

06

08

1

Input4

Deg

ree o

f mem

bers

hip


(d)


0 20 40 60 80 100 1200

01

02

03

04

05

06

07

Data index

Reco

very











0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very



Nomenclature




0

119894119895










component 119894



References



























VLSI Design



RotatingMachinery





Shock and Vibration



Advances in







Journal of





SensorsJournal of







Volume 2014

RoboticsJournal of





Journal of












01 02 03 04 05

005

01

015

02

025

03

035

04

045

05

Target = RF(exp)

Y=

RF(s

imul

ated

)


R = 09942

Data





0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very





ARE =

sum

119899

1

10038161003816100381610038161003816

(119881119900119881119900119894)sim minus (119881

119900119881119900119894)exp

10038161003816100381610038161003816

(119881119900119881119900119894)exp

119899

(7)




4 Conclusion



15 20 25 30 35

0

02

04

06

08

1

Input1

Deg

ree o

f mem

bers

hip


(a)

01 02 03 04

0

02

04

06

08

1

Input2

Deg

ree o

f mem

bers

hip


(b)

96 98 10 102 104

0

02

04

06

08

1

Input3

Deg

ree o

f mem

bers

hip


(c)

5 10 15 20

0

02

04

06

08

1

Input4

Deg

ree o

f mem

bers

hip


(d)


0 20 40 60 80 100 1200

01

02

03

04

05

06

07

Data index

Reco

very











0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very



Nomenclature




0

119894119895










component 119894



References



























VLSI Design



RotatingMachinery





Shock and Vibration



Advances in







Journal of





SensorsJournal of







Volume 2014

RoboticsJournal of





Journal of









15 20 25 30 35

0

02

04

06

08

1

Input1

Deg

ree o

f mem

bers

hip


(a)

01 02 03 04

0

02

04

06

08

1

Input2

Deg

ree o

f mem

bers

hip


(b)

96 98 10 102 104

0

02

04

06

08

1

Input3

Deg

ree o

f mem

bers

hip


(c)

5 10 15 20

0

02

04

06

08

1

Input4

Deg

ree o

f mem

bers

hip


(d)


0 20 40 60 80 100 1200

01

02

03

04

05

06

07

Data index

Reco

very











0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very



Nomenclature




0

119894119895










component 119894



References



























VLSI Design



RotatingMachinery





Shock and Vibration



Advances in







Journal of





SensorsJournal of







Volume 2014

RoboticsJournal of





Journal of











0 5 10 15 20 2502

02503

03504

04505

05506

065

Data index

Reco

very



Nomenclature




0

119894119895










component 119894



References



























VLSI Design



RotatingMachinery





Shock and Vibration



Advances in







Journal of





SensorsJournal of







Volume 2014

RoboticsJournal of





Journal of

















VLSI Design



RotatingMachinery





Shock and Vibration



Advances in







Journal of





SensorsJournal of







Volume 2014

RoboticsJournal of





Journal of








Modeling of Steam Distillation Mechanism during Seam Injection Process Using Artificial Intelligence

Documents

Transcript of Modeling of Steam Distillation Mechanism during Seam Injection Process Using Artificial Intelligence