Anda di halaman 1dari 8

GHTMDD - 337 Received: May 31, 1999

Accepted: October 10, 1999

Bulletin of the Chemists and Technologists of Macedonia,

Vol. 18, No. 2, pp. 171-178 (1999) ISSN 0350 - 0136 UDC: 665.637.4.001.26

Professional paper Dedicated to Professor Aleksandar Grizo on the occasion of his 80''' birthday

OPTIMIZA TION OF REFINERY

PRODUCTS

BLENDING*

Julija Risticl, Loreta Tripceva- Trajkovska1, Ice Rikaloskit, Liljana Markovska2 lOKTA Crude Oil Refinery, p.a. Box 66,91001 Skopje, Republic of Macedonia 2Department of Chemical and Control Engineering, Faculty of Technology and Metallurgy, The" Sv. Kiril & Metodij" University, 91000 Skopje, Republic of Macedonia

The oil products are manufactured by blending two or more different fractions whose quantities and physicochemical properties depend on the crude oil type, the way and conditions of processing. The quality of the oil products (fuels) for sale has to comply with the current standards for liquid fuels, and the produced quantities have to comply with the market needs.

It is in producer's interest to do the blending in an optimal way, namely, to satisfy the requirements for the oil products quality and quantity with a maximal usage of the available fractions and, of course, with a maximal profit out of the sold products.
The optimization of refinery products blending is accomplished lem is solved by using the spreadsheet solver WHAT'S BEST! Key words: oil products; blending; linear programming; optimization by applying linear programming. The prob-

INTRODUCTION Among the mathematical methods commonly usedby operations research analysts in modeling system;and organizations are the following: mathematical programming(linear, nonlinear, integer, dynamic, goal progranmring etc), network modeling, inventory modeling,queuing theory, game theory, simulation, forecastingand others [1].

Linear programming (LP) is the most often used method for solving optimization problems. Sincethe development of the simplex algorithm for solving linear programming problems by George Dantzig, LP has been used to solve optimization problemsin fields as diverse as petroleum, banking, educationand many others, and has a great importancein operations research methods [1, 2, 3].

Situations in which various inputs must be blendedin some desiredproportions to produce goods for sale are often amenable to linear programming analysis.Such problem')are calledblendingproblem'). The blendingproblem')are typical for the oil refining industrywhere various types of fractions and componentsare blendedto producedifferenttypes of gasoline and otheroil products. A useful tool for solving linear programming problems is WHAT'S BEST! software [4]. This software employs powerful optimization algorithms leading to dramatically reduced solution time of the optimization procedure when the problem contains a large number of variables and constraints.

* This paper was presented at the ZEMAK International Symposium "Production and Application of Oil Products" in Ohrid, September 30 - October 1,1999.

172

J. Risti,:, L. Trpc-eva-Trajkovska,

I. Rikaloski, L. Markovska

DEFINING THE PROBLEM

LP is an optimization problem of maximizing or minimizing the objective function which variables called decision variables must satisfy a set of linear constraints. The first step is to define the problem that has to be solved, which includes specifying the objectives that have to be achieved and the aspects that must be studied before the problem can be solved. The OKTA Crude Oil Refinery processes different types of crude oil to manufacturedifferentoil products for the market. These refinery products are manufacturedby blending various tractions produced by primary or secondary crude oil processing,which number, quantity and physical-chemical properties dependon the crude oil type, the way and conditionsof processing. The problem to solve is how to do the blending of the desired products in order to maximize the

sales revenues, while still maintaining an adequate. level of products quality and market supply. I The solution of the problem involves identifying a number of aspects that affect the achievement of the objectives: 1. Types of products needed by the market. 2. Products ex-refinery sales prices. 3. Minimal quality of the products that has to be achieved. 4. Market demand. 5. Available fractions to be blended into oil products. 6. Physical-chemical properties of all the fractions to be used as blending components. 7. Mathematical methods for estimating physical-chemical properties of the blended products.

DATA COLLECTING AND PROCESSING

After identifying the aspects that affect the problem, the next step is to collect the needed data and to estimate the values of all parameters necessary to develop and evaluate a mathematical model of the problem. The analysis of the main aspects, which could be evaluated, gives the following: 1. There are 8 products needed by the market that could be produced by the refinery: liquified petroleum gas (LPG), three types of motor gasoline - leaded regular (MB 86), leaded premium (MB 98) and unleaded premium (BMB 95), two types of diesel fuel (D1 and D2), extra light heating oil (EL) and fuel oil (FO) [5]. 2. The maximal retail as well as ex-refinery sales prices of oil products are under the authority of the government and they are the same all over the country. Some of the products are being sold by weight (LPG, FO) and some by volume (gasoline, diesel fuel, EL) [6] (Table 1). 3. The minimal quality of oil products is defined with the standards for liquid fuels that are currently in force in the country [6-10]. 4. The refinery product slate has to comply with the market demand, definedby the Energy balance of the country and the quantitiesof oil products suppliedon the domesticmarket in thepast [5, 11]. 5. As a result of the hydroskimming structure of the refinery and the current way of operation, there are 8 fractions available for blending: pro-

pane-rich fraction (F-C3), butane-rich fraction (FC4), light naphtha (LN), reformate (R), kerosene (K), light gas oil (LGO), heavy gas oil (HGO) and residual oil (RO). Their quantities could be derived from the material balance of the processing units. The reformate is the most important fraction for gasoline blending. Unleaded gasoline requires higher reformate octane quality and consequently, I different reforming unit operation conditions in comparison to leaded gasoline. So two different ways of operation of the reforming unit have to be defined and the material balance of all the processing units have to be related to the reformate quantity. A lot of data from our operating experience are at our disposal for this purpose [5].
I

Table

1
Oil products ex-refinery sales prices

Product
LPG MB 86 MB98 BMB 95 01 02 EK FO denlkg 13.167

Sales prices den/l 8.680 8.792 9.627 7.935 7.594 8.907 5.805

Bull. Chem. Technol. Macedonia, 18, 2,171-178

(1999) I

Optimization of refinery products blending

173

6. The physical-chemical properties of the fractionsused as blending components can be determinedeither experimentally or using empirical relations[5, 12, 13, 14]. The physical-chemicalpropertiesof the blending components are used to estimatethe specifications of the blended products. One of the assumptions that has to be satisfied for applyinglinear programming is the linear additivity of thephysical-chemical properties. This is the case withthe chemical composition, density, molecular weight,true vapor pressure (TVP), research octane number(RaN), sulphur content, cetane index, heat ofcombustionand TBP distillation. Some of the physical-chemical properties that don'tsatisfy the linear additivity assumption can be presentedwith blending indexes which have linear additivity characteristics. This is the case with kinematicviscosity, flash point, CFPP point and pour point. In the case of ASTM distillation, it has to be convertedfirst into TBP distillation. Having estimatedthe TBP distillation of the blend, it has to be convertedback to ASTM distillation that is speciTable 2

fied by the standards. The relation between the ASTM and TBP distillation is strictly empirical, and there is no way to predict the specified ASTM distillation point for the blended product (gasoline, diesel fuel) in advance, so an iterative procedure has to be applied. A similar iterative procedure applies for all physical-chemical properties which are functions of the distillation. For example, the standard for gasoline specifies RVP (reid vapor pressure), which also doesn't have linear additivity characteristics, so it has to be converted first into true vapor pressure (TVP). Having estimated the TVP of the blend, it has to be converted back to RVP. The relation between RVP and TVP is strictly empirical and depends on the tS%ASTM and tlS%ASTM points, which can't be predicted for the blended gasoline in advance. The same applies to the vapor-liquid ratio T36 that depends on the
t20%ASTM tsO%ASTM and points of the blended gasoline.

The physical-chemical propertie:;,; f the fraco tions used as blending components that have to be includedin the model are given in Tables 2-6.

Characteristics of LPG blending fractions Fraction


dlS g/m]
F-C)

c=1 M kglkmo] 58.09 44.27 [>38 kPa 393 1301 d1S g/m] 0.5785 0.5094

c=2
M kglkmo] 58.08 44.39 [>38 kPa 396 1293

0.5789 0.5087

F-C4

Table

3
Composition
c

characteristics of LPG blending fractions

=1
I:Cs 0.00 0.00 Ct+ C2 0.00 0.03

c=2
g (%wt) I:C4 99.78 2.84 I:Cs 0.00 0.00

Fraction
Ct+ C2 F-C) F-C4 0.00 0.04

g (%wt) I:C4 99.85 1.71

Table 4
Characteristics of gasoline blending fractions
c

=1
'jjJs'

c=2 RON
111.23 74.95 92.59
"'is"

Fraction
F-C4 LN R

'dis'" g/m] 0.5087 0.6337 0.7795

M
kglkmo] 44.27 74.07 1] 3.53

M kglkmo] 44.39 71.83 113.27

kPa 1301 III 480

g/m] 0.5094 0.6276 0.7817

'j.38 kPa ]293 127 52

RON

111.10 75.25 95.07

rJlac. xeM. TeXHon. MaKe)\OHHja, 18, 2,171-178

(1999)

174
Table 5.1

J. Risti<; L. Trp Ceva-Trajkovska, l. Rikaloski, L. Markovska

Characteristics Fraction K LGO HGO


dlS

of middle distillates
C.l.

blending fractions
VO 1111112/s 1.70 4.35 29.70

for c

=1
ICFPP C -48 20 10
Ipour

g/m! 0.802! 0.8360 0.8716

S %wt 0.043 0.20 0.57

Iflash c 58 90 138

DC -49 -21 9

42.63 50.08 51.08

Table

5.2
Characteristics of middle distillates
c.l.

blending fractions
VO mm2/s 1.95 4.79 34.67

for c

=2
IcFPP DC -42 -23 12 Ipour DC -43 -24 11

Fraction K LGO HGO

dlS g/m! 0.8098 0.8380 0.8752

S %wt 0.048 0.20 0.58

lfush C 64 90 128

43.69 50.76 50.52

Table

6 Characteristics of fuel oil blending fractions

Fraction

-----..-_uo-- c=1

c=2 Q MJ/kg 42.36 41.16


yloo
dlS

--dlS g/m! 0.8716 0.9457

S %wt 0.57 1.23

mm2/s 111.23 74.95

g/m! 0.8752 0.9410

s %wt 0.58 1.17

Q MJ/kg 42.30 41.24

yloo

mlllls 3.09 26.50

HGO Ra

7. For estimating some of the physicalchemical characteristics, blending indexes, ASTMTBP and RVP-TVP conversion, as well as for predicting the response of the blended gasoline to the antiknock compound (tetraethyllead), mathematical
FORMULATING

relations have been developed by means of correlation and regression analysis of experimental or published data. This is necessary in order to perform all the estimations by developing a spreadsheet model in EXCEL [1, 3, 5, 13, 14, 15, 16].
MODEL

THE LINEAR PROGRAMMING

The linear model should describe completely the decision we have to make: what quantity of what fraction should be blended in what product in order to maximize the sales revenues from the oil products, given the available crude oil and the restrictions from the current standards for liquid fuels, market demand and material balance of the processing units. The decision variables could be defined as follows:

Gcij - quantity of fraction i to be blended into

productj for operatingconditionsc, for i = 1,2, ..., 8;j = 1,2, ... , 8; c = 1,2, where c = 1 means operating conditions of the reforming unit for production of leaded gasoline, and c = 2 for unleaded gasoline. Taking in account the nature of the products that have to be blended and the accepted technology of production, the blending scheme could be presented in a matrix form as in Table 7. The presented schemedefines 39 decision variables.

Bull. Chem. Technol. Macedonia, 18,2,171-178

(1999)

Optimization of refinery products blending

175

Table

7
General blending scheme Product in matrix form 0)
5 Dl 6 D2 7 EL 8 FO
GlO

Fracti on

1 TNG

2 MB 86

3 MB98

4 BMB 95

Total

F-C3

1
2

GllI G2ll Gl2l G221 Gm G222 Gl32 G232 Gl42 Gl23 G223 Gm Gm GI43 G244 G155 G255 Gl65 G265 Gm G275 Gl56 G256 GI66 G266 Gl76 G276 Gl57 G257 G167 G267 Gm G277 Gl78 G278 GI88 G288 GO! G02 G03
GQ4 G05

F-C4

1 2

G20 G224 G30 G234 G4Q

LN

1 2

1 2

1 2

G50

LGO

1 2

G6Q

HGO

1 2

G70

RO

1 2

G80

Total

G06

GQ7

G08

Now the objective function (sales revenues) can be defined as follows: OBJFN = '[Ccij,gGcij

EmcijPciO ~ POj,max .

where: ccij,g - productsalespriceper weight. The values of the decision variables have to satisfy three sets of linear constraints, which have todescribe the limited resources or possibilities. The first set of constraints is defined by the requirements of the current standard for the minimal quality of the blended products and by the physical-chenlicalproperties of the available fractions. Different products have to satisfy different constraints. For example, the requirement for maximalvapor pressure can be defined as follows: 1. Maximal vapor pressure (LPG and gasalines)

In general, this form of inequality is not linear. But, after some arrangements, it can be converted into a suitable linear form:

E PciO - POj,max)

I M ciO)GciO~ O.

The coefficients of the decision variables in the constraints are technological coefficients. In this case the technological coefficient is:
(PciO - POj,max)M ciO'

In a similar way all the other constraints resulting from the quality requirements can be defined as: 2. Maximal content of component k (LPG)
'[(gciO,k -gOj,k,max)Gcij ~O.

rJJac. xeM. TeXHOJl. MaKenOHHja,

18, 2, 171-178

(1999)

176

J. Risti<~ L. Trp6eva-Trajkovska,

l. Rika/oski, L. Markovska

3. Minimal octane number (gasolines) I;"c(RONoj,min -RONciO)/dciO)Gcij sO.

I: (RI.v,ciO

- RI'v,Oj,max )Gcij sO.

4. Minimal quantity recovered by distillation (gasoline.,>nd diesel fuels) a

10. Minimal flash point - maximal blending index (middle distillates)

I: (RI.

flash,ciO - RI. flash,Oj,maxGcij sO. )

I: (( vbj,min -

v~iO) dciO)GcijsO. /

11. Maximal CFPP point - maximal blending index (diesel fuels)


I:B.I'CFPP ,ciO- B.J.CFPP,Oj,maxGcij sO. )

5. Minimal density (diesel fuels)


I:(1/dciO-1/dOj,min)Gcij sO.

6. Maximal density (diesel fuels) I:(l/doj,max -l/dcio)Gcij sO. 7. Maximal sulphur content (middledistillates)

I: (SciO -SOj,max

)Gcij sO.

8. Minimal cetane index (diesel fuels) I:(C.I'Oj,min -C.I'ciO)Gcij sO.

In total 43 constraints have to be defined for the quality requirements (LPG - 5, MB 86 - 5, MB 98 - 5, BMB 95 - 5, D1- 9, D2 - 9 andEL-5). The second set of constraints is defined by the market requirements. The restrictions of the market can be expressed as minimal or maximal quantity of some product, or as minimal or maximal content of some product in the pool. In this case 12 constraints are defined. The third set of constraints is defined by the material balance of the proce.,>sing nits, or in other u words, by the limited quantities of crude oil and fractions for blending. The quantity of crude oil and of the other seven fractions is related to the quantity of reformate for the two different operating situations - production of leaded and production of unleaded gasoline. So the total number of constraints for the material balance is 15. The model contains in total 39 variables and 70 constraints.

9. Minimal kinematic viscosity blending index (middle distillates)


I:(B.I'v,Oj,min

minimal

- B.I'v,ciO)Gcij sO.

10. Maximal kinematic viscosity - maximal blending index (middle distillates)

VERIFICATION OF THE MODEL

Having developed the linear programming model for solving the blending problem, it has to be determined how valid it is for the current situation. That's why the model has to be applied for solving real problems and predicting solutions for different alternatives. If the predicted values are close to the actual values, than the model is an accurate representation of the reality. The verification of the model has been performed for a case study of maximizing the sales revenues by processing 60000 tonnes of crude oil and production of winter quality oil products. The optimal solution was obtained using the WHAT'S BEST! software. The solution of the problem is given in Table 8. The resulting sales revenues are

511.4 million denars. The product range is close to the real values for the given crude oil type. The analysis of the optimal solution shows that every constraint of the model has been satisfied. The critical parameters are the minimal content of MB 86 and D1 and the maximal content of BMB 95 and EL in the respective pool, which shows that the prices of BMB 95 and EL are favorized and that it is unfavorable to produce MB 86 and Dl. Among the physical-chemical characteristics critical are the following: RON for BMB 95, vapor pressure for MB 98, distillation specifications for D1 and D2 and kinematics viscosity for EL. The other characteristics'show better quality than that specified by the current standards.

Bull. Ch em. Technol. Macedonia, 18, 2,171-178

(1999)

Optimization of refinery products blending

177

Table

8 Case study resulting blending scheme in matrix form Product0)

Fraction

I TNG

2 MB 86

3 MB98

4 BMB95

5 DI

6 D2

7 EL

8 FO

Total

F-C3

I 2

268 69 0 0 44 0 0 0 854 693 0 1141 186 7637 2253 0 984 2650 0 0 0 0 0 3365 1736 1206 0 4219 0 930 0 5169 782 0 1173 17736 4811 337 898 9658 2460 3634 6307 11437 23721 68 139

337

2 F-C4

1 2

877

3 LN.

I 2

1396

4 R

I 2

10744

5 K

1 2

5540

6 LGO

I 2

8681

7 HGO

1 2

8331

8 RO

1 2

22548

Total

58452

CONCLUSION

The linear programming method is very usefulfor solving blending problems. Similar linear progranmlingmodels can be applied in minimizing the variable costs or maximizing the profit, with different other objectives, like selecting the most suitablecrude oil types for processing for different periodsof the year and different market demands,

minimizing the product stocks etc. For a given alternative, sensitivity analysis may be performed for adjusting the processing conditions and analyzing the influence of various factors, like product and crude oil prices, products specifications, evaluating the benefit of a new product in the slate, reconstructions of theexistingunits and investmentin newunits.

REFERENCES
[I] W. L. Winston, Operations Research: Applications and Algorithms, Wads worth Publishing Company, Belmont, California, 1994. [2] S. Gass, Linear Programming, Third Ed., MC.Graw-Hill Book Company, New York, 1969. [3] R. Kiprijanova-Radovanovic, L. Markovska, Applied Mathematical Methods in Chemical Engineering, The "Sv. Kiril & Metodij" University, Skopje, 1999. [4] User's Manual, What's Best! The Spreadsheet UNDO Systems, Inc., Chicago, illinois, 1996. Solver,

fnae. xeM. TeXHOfl. MaKeAoHHja, 18, 2.171-178

(1999)

178
[5] OKTA Crude Oil Refinery, Commercial Documentation. [6] Standard for Propane-Butane

J. Risti(~ L. Trpeeva-Trajkovska,

I. Rikaloski, L. Markovska

Processing,

Research

and

Mixture MKS B.H2.134.

[11] Official Gazelle of the Republic of Macedonia, N 63/98, Regulation for the Energy Balance of the Republic of I Macedonia for 1999.

[7] Standard for Engine Gasolines MKS B.H2.220. [8] Qfficial Gazette of the Republic of Macedonia, N 13/86, Regulation for the Unleaded Motor Gasoline Quality. [9] Standard for Fuels for High Speed Compression-Ignition Engines MKS B.H2.41O. [10] Official Gazelle of the Republic of Macedonia, N 49/83, Regulation for the Fuel Oil Quality.

[12] P. Wuithier, Le petrole: Rafinage et genie chimique, Editions Technic, Paris, 1972. [13] Hemijanafte i naftnihproizvoda i procesi prerade u RNP, RNP, Pancevo, 1980.
[14] M. R. Riazi, T. E. Daubert, Predicting Flash and Pour Points, Hydrocarbon Processing, September 1987. [15] G. H. Unzelman et al., Octane Improvement Economics - Antinocks and Alternatives, 1979 NPRA Annual Meeting, San Antonio, Texas, 1979.

Pe311Me

OnTIIMAJlJf3AquJA

HA CMEIDYBAlbETO HA PAc])IIHEPIICKIITE npOII3BOD:II


TpHn'leBaTpajKoBcKal, HlI,e PHKaJIOCKH1, JIuJIjaua MapKoBcKa2

JYJIHja PHCTHKl, JIopeTa

lOKTA
211 flciIllllTiyiu

PaifJtmepllja
3a xeMllcKO

fW flacjJiIla, u. cjJax 66,91001


II KOfltTipOJlflO llH:J/Ceflepciuf3o,

Ciwuje,

Peuy6J1UKa MaKeooflllja
cjJaKYJliIletTi,

TeXfloJlOlllKO-MetUaJlyplllKll

Yflllf3ep3111UeiIl "Cf3. KllPllJl II M eiIlo 0 llj ", 91000 CKouje, Peuy6mtKa

MaKeooflllja

KJIY'Iuu 300pOBU: HaqJTeHH

npou3BoAH;

cMewYBa!:be;

nHHeapHo

nporpaMHpa!:be;

onTMMamBaIJ;Hja

HaqJTeHMTe

npoM3BoAM ce ].106UB<taT co CMeWYBa!:be M

HWIHH, O].1HOCHO].1a ce 3a].10BonaT

6apa!:baTa

BO norne)1

Ha ].1BeMnM nOBeKe pa3nWIHM qJpma.J;MM qMM KonWleCTBa HaqJTa H 0].1HaqMHOT MycnoBHTe KBanHTeTOT pHBaTa) Tpe6a Tpe6a ].1ap].1M 3a Bo Te'IHH HHTepec Ha npepa60TKa.

clJM3H'IKO-XeMHCKHOC06MHH 3aBHcaT 0].1THnOT Ha cypOBaTa Ha rOTOBHTe HaclJTeHH npOH3BOAH (roropMBa, a npoH3Be].1eHHTe KonH'IeCTBa

Ha KBanMTeTOT H KOnMqeCTBaTa Ha rOTOBHTe npOM3BOAH co MaKCMManHO HCKOpHCTYBa!:be Ha pacnonmKnHBHTe qJpaKIJ;HM M, ce pa36Mpa, ].1a].1eHMTenpOH3BO].1H. OnTMManM3aIJ;HjaTa apHO nporpaMHpa!:be. Ha CMeWYBa!:beTO e peweH BEST! Ha paqJHHeHa nHHe. co KopHCTeco MaKcHManeH npoclJHT 0].1 npo.

Aa 6HAe BO cornaCHOCT co B(l)Ke'IKHTe CTaHHa na3apOT. Ha onTHManeH

pMCKHTe npOH3BO].1M e HanpaBeHa TIp06neMoT !:be Ha coclJTBepCKHOT naKeT

co npHMeHa

Aa 6H].1aT BO cornaCHOCT co nOTpe6HTe Ha npOH3BO].1HTenOT npOM3BO].1H ].1a ce H3BpWH

e CMeWYBa!:beTO

WHAT'S

Ha rOTOBMTe

Bull. Chem. Technol. Macedonia, 18, 2,171-178

(1999)

Anda mungkin juga menyukai