Int. J. Oil, Gas and Coal Technology, Vol. 1, No.

3, 2008 283

A nonlinear programming model for refinery planning

and optimisation with rigorous process models and
product quality specifications

I. Alhajri and A. Elkamel*

Chemical Engineering Department,
University of Waterloo,
Waterloo, Ontario, Canada
E-mail: ialhajri@engmail.uwaterloo.ca
E-mail: aelkamel@uwaterloo.ca
*Corresponding author

T. Albahri
Chemical Engineering Department,
Kuwait University,
Safat, Kuwait
E-mail: albahri@kuc01.kuniv.edu.kw

P.L. Douglas
Chemical Engineering Department,
University of Waterloo,
Waterloo, Ontario, Canada
E-mail: pdouglas@uwaterloo.ca

Abstract: The yield of products in large-scale plants such as oil refineries have
a significant impact on overall profitability. Currently, many refineries apply
Linear Programming (LP) techniques for their production planning models.
However, this will often give inconsistent predictions of refinery productivity
and operation. Moreover, the stringent environmental regulations, product
qualities, and heavier feed stocks make it necessary to develop accurate models
for refinery-production planning. In this work, an approach with more accurate
representation of the refinery processes is presented. The resulting model is
able to predict the operating variables such as the Crude Distillation Unit
(CDU) cut-point temperatures and the conversion of the Fluid Catalytic
Cracking unit (FCC). It can also evaluate properties of the final products to
meet the market specification as well as the required product demands to
achieve a maximum refinery profit. The model is illustrated on representative
case studies, and the results are discussed. [Received: December 4, 2007;
Accepted: January 17, 2008]

Keywords: refinery planning; optimisation; product quality; blending.

Copyright © 2008 Inderscience Enterprises Ltd.

284 I. Alhajri et al.

Reference to this paper should be made as follows: Alhajri, I., Elkamel, A.,
Albahri, T. and Douglas, P.L. (2008) ‘A nonlinear programming model for
refinery planning and optimisation with rigorous process models and product
quality specifications’, Int. J. Oil, Gas and Coal Technology, Vol. 1, No. 3,
pp.283–307.

Biographical notes: Ibrahim Alhajri is a PhD candidate at the University

of Waterloo (Department of Chemical Engineering). He is currently working
in the area of planning and scheduling of refinery operations and hydrogen
management. He got his MS as well as his BS in Chemical Engineering from
Kuwait University.

Ali Elkamel is an Associate Professor of Chemical Engineering

at the University of Waterloo. Prior to joining the University of Waterloo,
he served at Purdue University, Procter and Gamble, Kuwait University, and
the University of Wisconsin. His research has focused on the applications
of systems engineering and optimisation techniques to pollution problems and
sustainable development.

Tareq Albahri is an Associate Professor of Chemical Engineering at Kuwait

University. He got his PhD from the University of Texas – Austin, and his BS
and MS from Kuwait University. His research interests are in the areas
of refinery systems engineering, refinery management, scheduling, and
operations research, petroleum characterisation, and computer-aided molecular
design and synthesis. Prior to going for his PhD at the University of Texas,
he served as a Process Engineer at Kuwait National Petroleum Company.

Peter L. Douglas is a Professor of Chemical Engineering at the University

of Waterloo. He specialises in process modelling, simulation, control and
optimisation. He has been a Visiting Professor at the University of Queensland
(Brisbane, Australia), University Teknologi Malaysia (Kuala Lumpur,
Malaysia), and King Mongkut’s University of Technology Thonburi
(Bangkok, Thailand). He has consulted on a worldwide basis for clients such as
DuPont, Shell, IBM, Goodyear, Alcan, CANMET, Government of Canada,
and National Science & Technology Development Agency of Thailand.

1 Introduction

Petroleum refineries extract and upgrade the valuable components of crude oil to produce
a variety of marketable petroleum products that are vital to everyday life. Examples
of these valuable products are gasoline, jet fuel and diesel. The petroleum-refining
industry employs a wide variety of processes as shown in Figure 1. It begins with the
Crude Distillation Unit (CDU) or fractionation of crude oils into separate hydrocarbon
groups. The resultant products are directly related to the characteristics of the crude
processed. Most distillation products are further converted into more usable products by
changing the size and structure of the hydrocarbon molecules through cracking,
reforming, and other conversion processes. Integrated refineries incorporate fractionation,
conversion, treatment, upgrading, and blending operations and may also include
petrochemical processing (Bodington, 1995; Favennec, 2001).
 A nonlinear programming model for refinery planning 285

Figure 1 Refinery flowchart of the illustrative case study

The aim in refinery operation is to generate as much profit as possible by converting

crude oils into valuable products. Mathematical programming or optimisation
has become an attractive tool to achieve this goal. Linear Programming (LP) has been
the most widely used technique in refinery planning and optimisation (Favennec, 2001).
Despite the many contributions that have been reported on planning models, very few can
be found that specifically address the petroleum-refining industry. Symonds (1956)
developed an LP model for solving a simplified gasoline refining and blending problem.
The advantage of LP is its quick convergence and ease of implementation. Allen (1971)
presented an LP model for a simple refinery that consists mainly of three units:
distillation, cracking and blending.
In reality, the actual refinery process is highly nonlinear. With the stringent
environmental regulations on the process industry to produce a salable product that meets
stricter product specifications while at the same time meeting environmental restrictions,
a more accurate model needs to be developed (Maiti, 2001). One of the first contributions
to consider nonlinearity in production planning is that of Moro et al. (1998).
The main objective of their study was to develop a nonlinear planning model for refinery
production. Their model is able to represent a general refinery topology and was tested
on a case study that dealt with the planning of diesel production. Pinto and Moro (2000)
also developed a nonlinear planning model for refinery processes. The model described
represents a general petroleum refinery, and its framework allows for the implementation
of nonlinear process models as well as blending relations. This model assumes
the existence of several processing units producing a variety of intermediate streams with
different properties that can be blended to constitute the desired products. However, some
refinery processes were modelled using simple linear relationships, which affect
the overall predictability of the overall model.
286 I. Alhajri et al.

Zhang and Zhu (2000) proposed a decomposition strategy to tackle large-scale

refinery optimisation problems. The approach is based on the resolution of the overall
refinery model into a master model and a number of submodels. The master model
determines interactions among the processes, while the submodels optimise individual
processes. The results from these submodels are fed back to the master model for further
optimisation. Later, Zhang et al. (2001) supplemented the refinery-planning model with
hydrogen and utility considerations. To make the problem more tractable, they adopted
a decomposition approach, in which material processing is optimised first using Linear
Programming (LP) techniques to maximise the overall profit. Then, the supporting
systems, which include the hydrogen network and the utility system, are optimised to
reduce operating costs for the fixed process conditions determined from the LP
optimisation.
Recently, Li et al. (2005) conducted a study on integrating crude distillation,
FCC and product blending modules into refinery-planning models. They presented
a refinery-planning model utilising simplified empirical nonlinear process models with
considerations for crude characteristics, product yields and qualities.
From the previous discussion, the need is clear to have an efficient refinery-planning
model with more accurate outcome for the petroleum refinery decision-maker. The model
should be capable to deal with different types of crudes without major changes
in the model. Also, the model should represent refinery operation planning to optimise
the operating variable in individual processing units. The most important operating
variable will be the CDU cut points, which will affect the product flow rates and
properties for all the streams in the refinery as well as the conversion in the other
processing units. The model should also meet market demand with quality constraints
for each final blended product. Nonlinear rigorous unit models will be used rather than
the linear models, which are based on yield vectors. A general model will embed
the different rigorous refinery process models and the blending model. Product qualities
as well as market demand will be taken into account.
The remainder of this paper is organised as follows: the problem statement is given
in Section 2, then the mathematical model of the general optimisation refinery planning
model and the processing unit models as well as the property blending models will be
presented in Section 3. In Section 4, different case studies will be used to illustrate
the overall refinery-planning model. Finally, conclusions from this research and future
work will be given in Section 5.

2 Problem statement

A petroleum refinery is an extremely complex entity. In general, a refinery is made

up of various processing units that separate crude oil into different fractions or cuts,
upgrade and purify some of these cuts, and convert heavy fractions into lighter ones.
The profitable operation of a refinery, therefore, requires the optimisation of different
intermediate and final products in addition to process feeds. Several trends in the oil
refinery industry are also leading to a tight production of different products with the new
more stringent specifications. The increased market for heavier crude oils, for instance,
forces refineries to increase their use of conversion units, where hydrocracking as a way
of upgrading heavy oils to more valuable products is employed (Elkamel et al., 1999).
This is in addition to the tightening of environmental regulations, which continuously
 A nonlinear programming model for refinery planning 287

reduce the allowed sulphur content in fuel products. A rigorous model of refinery
operations, which can capture the different refinery feed characteristics and which
mimics the different refinery stages more accurately, is attempted in this paper.
We consider an oil refinery that consists of several processing units, splitters, and
mixers. The final refinery products Vi,s, I = {refinery processing units} and s = {streams},
have to meet market demand and specification. The processing units have operating
variables XUi,x, x = {operating variables} that affect the product flow rates and properties.
The different units are connected by splitters and mixers. The overall objective
is to maximise the refinery profit by adjusting the flow rates of different streams,
intermediates or final products, as well as the operating variables for the processing units.

3 Mathematical model

An individual nonlinear mathematical model is developed for each unit shown

in the refinery (Figure 1) prior to the development of a planning model, the connections
between the streams, and the blending pool. The processing units in this work are
modelled using nonlinear regression correlations. The correlations are developed
for CDU, hydrotreaters, reformer, FCC and Hydrocracker (HC). Our final aim
is to provide methods of determining optimal operational plans for a petroleum refinery
including the fraction cut points of the CDU and the severity or conversion
of the processing units. The most important variables in operational planning models are
the processing unit operating variables, feed flow rates, feed properties, product flow
rates, and product properties.
The mathematical formulation and representation in this study will follow closely that
of Neiro and Pinto (2005) with the improvements mentioned earlier. The objective
function will be the maximisation of the total profit of the refinery, i.e.,
Max profit = ∑ Cpi Fi − ∑ Cfi Fi − ∑ Cxi Fi . (1)
i∈B i∈E i∈I

Equation (1) expresses the overall refinery profit as revenues from selling all products,
subtracting costs of purchasing feedstock and costs of operating process units
in the refinery. B represents the set of blending units for the final products and their
sales price Cpi. The cost Cfi of the feedstock purchased from external sources defined
under the set (E) for all the units that receive such material from outside. Finally, there
is an operating cost Cxi for each processing unit (i) in the refinery where it is usually
expressed as a function of the quantity fed to the unit.

3.1 General model

A generic processing unit drawing is shown in Figure 2 to illustrate the mathematical
representation. The general mathematical model consists of the following sets
of constraints:
288 I. Alhajri et al.

Figure 2 General processing unit model

• Feed flow rate of processing units:

Fi = ∑ j∈J ∑ s∈N VS j ,s ,i ∀ i ∈ I . (2)

The feed Fi for any processing unit (i ∈ I, I is the defined set of all the units in the
refinery) is the summation of all flow rates VSj,s,i of the possible streams s that can be
received by unit (i) from units (j ∈ J), where J is defined as the set of all units that can
send streams to unit (i) and N is defined as the set of all streams s that can be sent from
unit (j) to unit (i).
• Feed properties of processing units:
FPi ,p = f (VS j ,s ,i , PV j ,s ,p ) ∀ i ∈ I , p ∈ PFi . (3)

Properties (p) of the feed to unit (i) are represented by FPi,p, and PFi is the set of all feed
properties to unit (i). The properties are functions of the quantities and properties
of all streams s from unit (j), VSj,s,i, and PVj,s,p, respectively. For example, the sulphur
weight percent on the catalytic reformer unit feed is written as:

FPCR ,SUL =
( ∑ VSj j ,Naph, CR × PV j ,Naph, SUL ).
( FCR × FPCR , SG )

• Product flow rates of processing units are given as:

Vi ,s = f ( Fi , FPi ,p , XU i ,x ) ∀ i ∈ I , s ∈ Si , x ∈ X (4)

The product flow rate from unit (i) for stream (s) represented by Vi,s (s ∈Si; Si is the
defined set of all the streams produced from unit i) is a function of the unit (i) feed
quantity Fi and property FPi,p as well as the operating variables XUi,x (x ∈ X; X is the
defined set of all operating variables). For example, the light naphtha produced from the
Fluid Catalytic Cracking (FCC) unit is written as:
 A nonlinear programming model for refinery planning 289

VFCC,LN = FFCC × FPFCC, SG ×  ∑ h = 0 (ah + bh × FPFCC,K ) × XU FCC,CONV% 

4

 
where ah and bh are constants.
• Product properties of processing units:
PVi ,s ,p = f ( FPi ,p , XU i ,x ) ∀ i ∈ I , s ∈ Si , p ∈ Pi (5)

PVi,s,p is the product property (p) for product stream (s) from unit (i), which is a function
of unit (i) feed properties FPi,p and the operating variables XUi,x. For example, the flash
point temperature of the kerosene produced by the HC unit is written as:

( FP + 460) ^ 0.333 
PVHC ,Kero,FLSH = 120 − 4 ×  HC ,VABP  − [0.05 × XU HC ,CONV% ]
 FPHC ,SG 

• Product splitting:
Vi ,s = ∑ m∈M VSi ,s ,m ∀ i ∈ I , s ∈ Si . (6)

The above equation represents the possibility for each product from unit (i) to be split
into many streams either as final product or feed to other processing units. Product stream
(s) from unit (i) represented by Vi,s can be sent to different destinations (m) defined
by streams VSi,s,m (m ∈ M; M is defined as the set of all the possible units or final
products pool blending that can receive the splitted streams).
• Processing unit capacity:
Fi ≤ Umax i ∀ i ∈ I. (7)

The feed of processing unit (i) cannot exceed its maximum capacity, which is represented
by Umaxi.
Equations (2) and (3) represent the feed quantities and properties of the processing
unit models, which play an important role in the product flow rates and properties,
defined by equations (4) and (5). Clearly, equations (2), (6) and (7) are linear whereas
equations (3)–(5) are nonlinear owing to mixing.

3.2 CDU model

The crude oil should be characterised before being fed to the CDU. One of the key
attributes for characterising the hydrocarbons composing crude oils is by boiling point.
This attribute is determined through laboratory test methods by measuring
the temperature at which the components of the crude oil will evaporate at a given
pressure (typically atmospheric pressure unless stated to be a different pressure basis).
A True Boiling Point (TBP) curve is developed as a part of the crude assay to determine
the liquid volume percent of the crude oil that evaporates relative to temperature
at atmospheric pressure (Watkins, 1979; Maples, 1993; Gary and Handwerk, 1994).
Figure 3 shows the TBP curve for the crude assay (Alaska) used in this study. The cuts
produced in the CDU are shown in Table 1.
290 I. Alhajri et al.

Figure 3 TBP distillation curve (Crude: Alaska) (see online version for colours)

Table 1 Boiling range of typical crude oil fractions

Fraction TBP – Boiling range (F)

SRLN 90–220
SRHN 180–380
Kerosene 330–520
Diesel 420–630
VGO 610–1050
Residue 950+

The mathematical model for the CDU is expressed by constraints similar to the general
constraints (2)–(7) discussed earlier. The same notation will be used here, where the unit i
for this case will be the CDU unit. The operating variable of the CDU unit is the cut-point
temperature for fraction (s), x = TECDU. Also, the products stream for the CDU unit are
fractions s (s ∈ SCDU = LPG, SRLN, SRHN, Kero, Diesel, VGO, and Rsd). The CDU
model is described as follows:
4
Cuts = ∑ ak (TECDU ,s ) k ∀ s ∈ SCDU − {Rsd } (8)
k =0

Cuts represents the volume percent vapourised of all fractions (s), except the residue
product, of CDU unit. The cuts are usually represented as a polynomial function in
TECDU,s, which is equivalent to the End-Point Temperatures (EP). For every product from
the CDU, the TECDU,s has an upper and a lower bound, which is called the swing cut.
Figure 4 shows an illustration of the CDU cuts volume as a function of the fractions
temperature TECDU,s. The coefficients of the polynomial of the CDU equation are listed in
Table 2. The residual cut volume percent is expressed as:
CutCDU ,s = Rsd = 100. (9)
 A nonlinear programming model for refinery planning 291

Figure 4 Swing cut of CDU fractions

Table 2 CDU unit model equations coefficients

Cut % (Vol.)
Parameter equation (8) API equation (12) SUL% equation (12) N% equation (12)
a0 4.040637061 81.84796736 0.050579083 –0.000882902
a1 –0.047271899 –3.778147973 –0.02036269 0.000304355
a2 0.000324992 0.113288448 0.001849373 –2.2968E–05
a3 –2.84324E–07 –0.0015436414 –3.25656E–05 4.58852E–07
a4 8.15312E–11 7.19024E–06 2.0301E–07 6.76957E–09

Since the last cut is the residue of the crude, it will be assumed that the accumulative
vapourised percent will be 100%.
Each product volumetric flow rate is calculated by subtracting its accumulated
volume percent vapourised from the previous cut volume and multiply the result with
crude feed to the CDU, i.e.,
 Cuts − Cuts −1 
VCDU ,s = FCDU ×   ∀ s ∈ SCDU (10)
 100 

VCDU,s represents the volume flow rate of all the products (s) from the CDU unit, and
FCDU is the crude oil to the CDU.
Properties of each product from the CDU (API, sulphur, etc.) are expressed as
polynomial functions in each product mid-volume percent vapourised. The mid-volume
for any product can be calculated from averaging the accumulative current cut volume
percent with the previous cut volume percent vapourised:
Cuts + Cuts −1
MidVCDU ,s = ∀ s ∈ SCDU (11)
2
292 I. Alhajri et al.

4
PVCDU ,s ,p = ∑ ak MidVs
k
∀ s ∈ SCDU , p ∈ Ps (12)
k =0

PVCDU,s,p represents different properties (p) for each product (s) from the CDU unit.
Ps is the set of all the properties calculated for the specified stream (s).
VCDU ,s = ∑ VS
m∈M
CDU ,s ,m ∀ s ∈ SCDU (13)

VSCDU,s,m represents the volume flow rate of all the streams split from the CDU
products (s) to different destinations (m), as explained in equation (6).
All fractions for the CDU, except residue, have an upper and a lower limit for their
cut-point, i.e.,
L
TECDU U
,s ≤ TECDU ,s ≤ TECDU ,s ∀ s ∈ SCDU − {Rsd }. (14)

Also, the crude feed to the unit cannot exceed its throughput capacity:
FCDU ≤ Umax CDU . (15)

3.3 Processing unit models

The other processing units have different operating variables. The aim of the refinery-
processing unit models is to predict and evaluate feedstocks and operating conditions to
plan the refining operations. In this study, simplified nonlinear process correlations are
used to predict product yields and properties for every processing unit. Table 3 lists the
processing units feed and products as well their properties. The operating variables for
the different processing units are listed in Table 4. In the next subsection, we present the
FCC unit model as an illustration of the processing unit models.

Table 3 Processing units feed and product streams and properties

Unit Feed streams Feed properties Products streams Products properties

Reformer Naphtha HT’s ARO% NAPH% Reformate RON
HNHC RVP
FCC TGO VABP LNFCC RON, RVP
AP HNFCC RON, RVP
Kf LCOFCC CI, FLSH
HCOFCC VISCO
HC TGO Kf LNHC RON, RVP
HNHC RON, RVP
KHC FLSH, FRZ
DHC CI, FLSH
NHT HNSR N%, Metal TLN ARO%, NAPH%
THN ARO%, NAPH%
DHT DSR N%, Metal NDHT ARO%, NAPH%
KDHT FLSH, FRZ
TDiesel CI, FLSH
 A nonlinear programming model for refinery planning 293

Table 3 Processing units feed and product streams and properties (continued)

Unit Feed streams Feed properties Products streams Products properties

GOHT VGO N%, Metal NGOHT ARO%, NAPH%
DGOHT CI, FLSH
TGO VABP, AP
RDHT RSD N%, Metal NRDHT ARO%, NAPH%
DRDHT CI, FLSH, VISCO
TRSD VISCO
*API and S% are required in all streams.

Table 4 Operation variables used for predicting product yields and properties

Processing unit Operating variable

Catalytic Reforming (Reformer) Reformate octane number
Fluid Catalytic Cracking (FCC) Conversion
Hydrocracking (HC) Conversion
Hydrotreating (HT) Conversion

3.3.1 FCC unit model

FCC is the most widely used catalytic cracking process. Many refiners call the FCC
the heart of the refinery (Maples, 1993). The products stream are Light Naphtha Gasoline
(LN), Heavy Naphtha Gasoline (HN), Light Catalytic Gas Oil (LGO) and Heavy
Catalytic Gas Oil (HGO). There have been several efforts at modelling the FCC unit and
other refinery-processing units. Various correlations in the literature for predicting
the FCC product yields and properties have been published. Al-Enezi et al. (1999)
presented linear and nonlinear regression models for predicting yields and properties
of the FCC process and tested them against refinery data. The feedstock properties and
the conversion were used as the correlation parameters. Li et al. (2005) proposed
a nonlinear correlation for predicting the yields without considering the property
correlations. The FCC model we employed is that of Al-Enezi et al. (1999). It consists
of the following equations, where all the parameter values are listed in Table 5.

Table 5 FCC unit model equations coeffecients

Equation Parameter LN HN LCO HCO

Product yield a0 –0.00308535 –0.001778704 0.0325 0.9675
a1 0.001337565 0.000771106 0.00175 –0.01175
a2 4.34839E–05 2.50685E–05 –0.00005 0.00005
a3 –1.29355E–07 –7.45732E–08 0 0
a4 –6.5E–10 –3.74725E–10 0 0
b0 0.02374645 0.013689828 0 0
b1 0.000104739 6.0382E–05 0 0
b2 –1.76439E–05 –1.01717E–05 0 0
b3 1.54805E–07 8.92451E–08 0 0
b4 6.5E–10 3.74725E–10 0 0
294 I. Alhajri et al.

Table 5 FCC unit model equations coeffecients (continued)

Equation Parameter LN HN LCO HCO

API a0 1672 7427 35 17
a1 70 –41 –0.2 –0.2
b0 0 654 5 6
b1 0 –6 0 0
c 30.7 169.3 1 1
d 1 –1 0 0
SUL% a 2.215 17.745 1.3 2.5
b 0.01 –0.12 0 0
c 30.7 169.3 1 1
d 1 –1 0 0
RON a 2906.1 15417.9
b 90 –90
c 30.7 169.3
d 1 –1
FLSH a 81750
b –750
c 650
d 35
e –1

• Products yield:

 4 
V FCC, s = F FCC × FP FCC, SG ×  ∑ ( a s , h + b s , h × FP FCC, K ) × XU FCC,Conv%  s ∈ SFCC
h
(16)
 h=0 
VFCC,s represents the yield for FCC unit products, where FPK is the characterisation factor
of the FCC feed and the operating variable x is the conversion of the FCC unit.
• Products properties:
The products APIs and sulphur content are a function of both the characterisation factor
of the feed and conversion of the unit, i.e.,
 1 
 ∑ (as , h + b s , h × FP FCC, K ) × XU FCC,Conv% 
h

PVFCC,s ,API =  h=0  s ∈ S FCC (17)

(cs + d s × XU FCC,Conv%)

and
[(as + b s × XU FCC,Conv%)]
FP FCC,SUL% × s ∈ S FCC (18)
(c s + d s × XU FCC,Conv%)

RON (equation (19)) and the flash point (equation (20)) were correlated as function of
LN and HN:
 A nonlinear programming model for refinery planning 295

(as + b s × XU FCC,Conv%)
PVFCC, s ,RON = s = LN , HN ∈ S FCC (19)
(cs + d s × XU FCC,Conv%)

[(as + b s × XU FCC,Conv%)]
PV FCC, s ,FLSH = 210 +
(cs + d s × XU FCC,Conv% + e s × XU FCC,Conv%
2
) (20)
s = Kero ∈ SFCC .

3.4 Properties blending

Refinery products are typically the result of blending several components or streams.
The purpose of the blending process is to obtain petroleum products from refined
components that meet certain quality specifications. Increased operating flexibility and
profits result when refinery operations produce basic intermediate streams that can be
blended to produce a variety of on-specification finished products. In this study, several
blending properties are included in the general model. Blending indices for each property
are used throughout the paper. The method of finding any blending property is via
finding the blending indices (PI) for each stream by a property equation for a given
property (p) to be blended and averaged, the Blending Index (BI) can be expressed
by the following general equation:
BI p = ∑ s PI s ,p × ys ,p ∀ p∈P (21)

where BIp represents the BI for a property p. PIs is the property index for the property p
of a stream s and ys is either weight or volume fraction depending on the property.
The properties covered in this study are given below.

3.4.1 API
The density of petroleum oil is expressed in terms of API gravity rather than specific
gravity. The blended API can be calculated by the following equation (Gary and
Handwerk, 1994):
141.5
API blend = − 131.5. (22)
SGblend

Specific Gravity (SG) can be averaged while API cannot. Therefore, the SG of the blend
can be calculated as:

SGblend =
∑ V × SG
s s s
(23)
∑V s s

Vs represents the volume percent of stream s and SGs is the specific gravity of stream s.

3.4.2 Sulphur content (wt%)

Sulphur content is an important property, which has a major influence on the value
of crude oil and petroleum products. The sulphur content for a blended stream SULblend
is the average sulphur content for all coming streams SULs and should be expressed
296 I. Alhajri et al.

in weight percent. It can be calculated from the following equation (Gary and
Handwerk, 1994):

SUL blend =
∑ W × SUL
s s s
(24)
∑W s

where Ws is the weight flow rate for stream s being blended.

3.4.3 Octane Number (ON)

Octane numbers are blended on a volumetric basis using the BI of the components.
True octane numbers do not blend linearly and it is necessary to use blending octane
numbers in making calculations. Several blending approaches are provided in the
literature and the simplest form has been converted to the following analytical relation
(Riazi, 2005):
RONIs = 651 z 3 − 1552.9 z 2 + 1272 ( z ) − 299.5 76 ≤ RON ≤ 103 (25)

where z = RON/100

RON blend =
∑ V × RONI
s s s
. (26)
∑V s s

3.4.4 Reid Vapor Pressure (RVP)

The Reid Vapour Pressure (RVP) is one of the important properties of gasoline and jet
fuels, and it is used as a criterion for blending products. RVP is the absolute pressure
exerted by a mixture at 100F. The approach for calculation of RVP of a blend when
several components with different RVPs are blended is often to use a BI for RVP as
(Riazi, 2005):
RVPBI s = RVPs1.25 (27)

RVPBI blend =
∑V s s × RVPBIs
(28)
∑V s s

RVPblend = [RVPBI blend ]0.8 (29)

where RVPBIs is the BI for RVPs. RVP can be calculated in bar or psia in the above
equation. This relation was originally developed by Chevron and is also recommended in
other industrial manuals under Chevron blending number.

3.4.5 Flash point

Flash point is an important characterisation of light petroleum fractions and products
under a high temperature environment and is directly related to the safe storage and
handling of such petroleum products. The flash point of the blend should be determined
from the flash point indexes of the components as given by (Riazi, 2005):
 A nonlinear programming model for refinery planning 297

2414
log10 FLSHI s = −6.1188 + . (30)
FLSH s − 42.6

FLSHIS is the flash point BI of stream s and FLSHs is the flash point
in degrees Kelvin. The blend flash point index can be determined from the general
equation (21) with a volume averaging.

3.4.6 Cetane index

For diesel engines, the fuel must have a characteristic that favours auto-ignition.
The ignition delay period can be evaluated by a fuel characterisation factor called Cetane
Number (CN). The Cetane Index (CI) is empirically correlated to the API gravity and
Aniline Point (AP) in °C. CI is expressed as follows (Riazi, 2005):
 ( API s )(1.8 APs + 32) 
CI s = 0.72 ×   + 10. (31)
 100 
The BI is calculated with volume fraction as in equation (21).

3.4.7 Freezing point

Freezing point is one of the important characterisations of aviation fuels. The equation
to calculate the freezing point index is (Baird, 1987):
 FRZs + 460 
FRZI s = Exp  13.333ln  (32)
 600 
where FRZs represents the freezing point of stream s in (F). The BI is calculated with
volume fraction as in equation (21).

4 Case studies

To illustrate the model of the previous section, different case studies are considered.
Figure 1 shows a simplified flow diagram for the oil refinery under consideration.
A single or mixture of crude oils can be charged to the CDU unit. CDU consists
of an atmospheric and vacuum distillation tower. Different fractions are then withdrawn
from the unit including Straight Run Light Naphtha (SRLN), Straight Run Heavy
Naphtha (SRHN), Kerosene (Kero), diesel, Vacuum Gas Oil (VGO) and residue.
The overhead gases are sent directly to a Liquefied Petroleum Gas (LPG) plant.
The hydrotreating is utilised to remove the sulphur from the intermediate streams.
The SRLN stream from the top of the distillation column is sent to a gasoline pool for
blending. The SRHN stream from the CDU after being hydrotreated in NHT is fed
to a catalytic reformer. The Catalytic Reforming (CR) process reforms the molecular
structure of the heavy naphtha to increase the percentage of high-octane (for gasoline
blending). The VGO stream, after being hydrotreated in GOHT, is fed to the FCC unit
and the HC unit.
298 I. Alhajri et al.

The FCC process converts heavy gas oils into lighter products, which are then used as
blendstocks for gasoline and diesel fuels. The HC unit is similar to the FCC unit
to the extent that this process catalytically cracks the heavy molecules that comprise gas
oils by splitting them into smaller molecules, which boil in the gasoline, kerosene, and
diesel fuel boiling ranges. The residue from the bottom of the distillation column
is hydrotreated and then directed to fuel oil pool.
The model was implemented in the General Algebraic Modelling System (GAMS)
Brooke et al. (1998) and solved with the CONOPT solver (Drud, 1994). Different initial
starting points were used and the best solution was retained. The model optimises
all intermediate and final product streams across the oil refinery subject to connectivity,
capacity, demand, and quality constraints. These constraints can be easily modified
to either include new data or guide the model to acceptable solutions.

4.1 Base case study

The refinery for this base case study imports crude oil and MTBE to produce five final
products. The crude oil feed to the refinery is assumed to be constant with
a 1,00,000 bbl/day Alaska crude oil (49.26 $US/bbl). The MTBE (50.20 $US/bbl) used
for improving the octane number of the gasoline pool (Energy Information
Administration, 2006). The refinery has to meet the market demand for different products
(Table 6) as well as the product specifications (Table 7). The objective is to maximise the
overall refinery revenue while meeting both market demand (in terms of both quantity
and quality). A summary of each blending pool final product stream flow rates and
properties is given in Table 8. The table shows the solution of the base case study.
The optimisation procedure leads to the production of profitable products and to meet
all product demands and quality restrictions. The intermediate products were blended first
to meet the model constraints and to maximise the overall refinery profit. For example,
the intermediate streams are sent to a less profitable blending pool rather than profitable
ones. Also, the distillate produced by the diesel hydrotreater (KDHT) is chosen to be
blended with the LSFO pool rather than kerosene or diesel, the reason for this
is the property specification limitation on the LSFO viscosity. On the other hand, when
the quality constraints are met, the model opts to send the streams to the profitable pool.
For example, in the case of DGOHT and DRDHT (5090 and 1320 bbl/d, respectively),
the model recommends sending these streams to the LSDSL pool instead of the
LSFO pool.

Table 6 Products demand

Final product Demand Bbl/day Price $US/Bbl

PRG 15,000 75.14
RGG 15,000 71.73
ATK (Jet Fuel) 15,000 73.79
LSDSL 15,000 72.99
LSFO 15,000 46.59
 A nonlinear programming model for refinery planning 299

Table 7 Products specifications

Final product Property Specification requirement

PRG API ≥45.0
SUL% ≤0.05
RON ≥92.0
RVP, psi ≤8.8
OXG% ≤2.2
RGG API ≥45.0
SUL% ≤0.05
RON ≥89.0
RVP, psi ≤8.8
OXG% ≤2.2
ATK API 37.0 ≤ API ≤ 51.0
SUL% ≤0.30
FLSH, °F ≥130
FRZ, °F ≤–40
LSDSL API ≥35.0
SUL% ≤0.05
CI ≥45
FLSH, °F ≥100
LSFO API ≥10
SUL% ≤1.0
VISC, cSt ≤150

Table 8 Blending products flow rate and properties

Blending Flow rate Final product Product Property

pool Product stream (Bbl/day) (Bbl/day) property value
PRG SRLN 2,530 15,000 – –
TLN 1,260 API 45.0
REFORMATE 8,825 SUL% 0.003
LNHC 0 RON 92.8
LNFCC 15 RVP 8.8
HNFCC 250 OXG% 2.0
C4 455 – –
MTBE 1,665 – –
RGG SRLN 3,085 15,000 – –
TLN 0 API 56.9
REFORMATE 3,780 SUL% 0.002
LNHC 2,000 RON 90.2
LNFCC 2,520 RVP 8.8
HNFCC 1,670 OXG% 2.0
C4 315
MTBE 1,670 – – –
300 I. Alhajri et al.

Table 8 Blending products flow rate and properties (continued)

Blending Flow rate Final product Product Property

pool Product stream (Bbl/day) (Bbl/day) property value
ATK Kero (CDU) 12,740 25,975 API 40.4
KHC 13,235 SUL% 0.145
KDHT 0 FLSH 166.9
– – FRZ –42.7
LSDSL TDiesel 8,815 29,030 – –
DHC 13,800 API 35.0
LCOFCC 0 SUL% 0.034
KDHT 0 CI 57.4
DGOHT 5,090 FLSH 217.5
DRDHT 1,320 – –
LSFO TRSD 15400 18,980 – –
LCOFCC 1,055 API 10.7
HCOFCC 370 SUL% 0.198
KDHT 2,155 VISC 150.0
DGHOT 0 – –
DRDHT 0 – –

4.2 Maximum products case study

In this case, we consider the maximisation of production in anticipation of increased
market demand. The quality specification constraints are still enforced. Four different
cases are presented each dealing with the maximisation of a specific product
(gasoline, ATK, LSD, and LSFO). Tables 9–12 show the final product flow rates and the
quality specifications for each product in each case.
In the gasoline case (Table 9), the objective was to maximise both PRG and RGG
with a minimum of 5000 BPD each. The optimisation results show an increase in the
RGG rather than the PRG owing to the lower RON value of the RGG and because
maximising the profit is not any more the objective. Also, we can see clearly that there is
a quality give-away on the RON for RGG. Moreover, from this case, we can see the
importance of the nonlinear model in meeting the properties specification as depicted by
the operating variables in the different refinery processes. For example, when maximising
gasoline, the expected result was to run the FCC with full throughput rather than running
the HC. However, the bottleneck was the ATK freezing point, which forced the model to
select to run HC rather than a production rate of 5990 BPD.

Table 9 Maximum gasoline case study

Final product Flow rate (Bbl/day) Product property Property value

PRG 8,880 API 47.4
SUL% 0.05
RON 92.8
RVP 8.8
OXG% 2.0
 A nonlinear programming model for refinery planning 301

Table 9 Maximum gasoline case study (continued)

Final product Flow rate (Bbl/day) Product property Property value

RGG 34,670 API 55.2
SUL% 0.003
RON 92.0
RVP 8.8
OXG% 2.0
ATK 6,740 API 41.5
SUL% 0.107
FLSH 167.2
FRZ –40.0
LSDSL 27,810 API 35.6
SUL% 0.032
CN 56.1
FLSH 222.8
LSFO 22,800 API 11.0
SUL% 0.18
VISC 49.9

For the ATK and LSDSL case studies (Tables 10 and 11, respectively), the property
constraints for the gasoline (API, SUL, and RON) and the viscosity of the LSFO were the
binding constraints. For the LSFO case (Table 12), the lighter product specifications were
the binding constraints.

Table 10 Maximum ATK case study

Product Flow rate (Bbl/day) Product property Property value

PRG 15,490 API 47.4
SUL% 0.05
RON 92.0
RVP 6.88
OXG% 2.0
RGG 5,000 API 61.8
SUL% 0.0005
RON 89.0
RVP 8.24
OXG% 2.0
ATK 37,045 API 40.6
SUL% 0.131
FLSH 167.6
FRZ –40.7
302 I. Alhajri et al.

Table 10 Maximum ATK case study (continued)

Product Flow rate (Bbl/day) Product property Property value

LSDSL 26,515 API 35.2
SUL% 0.035
CN 59.0
FLSH 211.0
LSFO 20,085 API 12.4
SUL% 0.183
VISC 150.0

Table 11 Maximum LSD case study

Product Flow rate (Bbl/day) Product property Property value

PRG 18,855 API 45.0
SUL% 0.05
RON 92.8
RVP 6.88
OXG% 2.0
RGG 5,000 API 60.2
SUL% 0.0002
RON 89.0
RVP 8.19
OXG% 2.0
ATK 20,470 API 43.0
SUL% 0.048
FLSH 165.9
FRZ –52.3
LSDSL 41,180 API 35.2
SUL% 0.034
CN 57.7
FLSH 218.5
LSFO 18,585 API 11.7
SUL% 0.197
VISC 150.0

4.3 Maximum profit with free demand case study

This case study was undertaken to determine the production plan of the refinery in terms
of intermediate and final products to achieve the maximum possible profit for the refinery
without paying attention to market demand. It is implicitly assumed here that the refinery
is able to sell all what it can produce. A minimum production rate of 5000 BPD
was imposed on both PRG and RGG. The results (Table 13) show clearly a focus
 A nonlinear programming model for refinery planning 303

on production of LSDSL and ATK rather than gasoline. Again the viscosity restriction
on LSFO forced the model to produce a large quantity of LSFO. The profit is improved
in this case by a margin of 2.3% compared with the base case.

Table 12 Maximum LSFO case study

Product Flow rate (Bbl/day) Product property Property value

PRG 25,415 API 45.0
SUL% 0.05
RON 92.8
RVP 6.19
OXG% 2.0
RGG 10,405 API 65.3
SUL% 0.004
RON 90.9
RVP 8.8
OXG% 1.7
ATK 6,740 API 41.5
SUL% 0.108
FLSH 167.2
FRZ –40.0
LSDSL 18,885 API 34.2
SUL% 0.04
CN 57.4
FLSH 233.5
LSFO 37,820 API 15.4
SUL% 0.131
VISC 21.7

Table 13 Maximum profit (free demand) case study

Product Flow rate (Bbl/day) Product property Property value

PRG 5,000 API 48.1
SUL% 0.001
RON 92
RVP 8.8
OXG% 2.0
RGG 16,280 API 54.2
SUL% 0
RON 89.6
RVP 8.8
OXG% 2.0
ATK 28,405 API 42.3
SUL% 0.077
FLSH 166.6
FRZ –45.1
304 I. Alhajri et al.

Table 13 Maximum profit (free demand) case study (continued)

Product Flow rate (Bbl/day) Product property Property value

LSDSL 36,620 API 35.2
SUL% 0.034
CN 57.9
FLSH 217.7
LSFO 18,585 API 11.6
SUL% 0.197
VISC 150.0

Table 14 provides a comparative summary of the different case studies considered.

The optimal temperature cut for the CDU fractions and the optimal FCC conversion
are different for the different case studies. It is important to note that the model was able
in each case to provide the best operational capacity of the refinery-processing units.
The optimisation chose to shutdown the FCC unit in the maximisation case studies
of ATK, LSDSL, and free market demand.

Table 14 Comparison between selected variables for different case studies

Max profit
Base case Max gasoline Max ATK Max LSDSL Max LSFO (free demand)
CDU cut point (F)
SRLN 220.0 220.0 194.1 208.7 208.8 213.8
SRHN 380.0 380.0 330.0 369.5 380.0 330.0
KERO 506.1 420.0 520.0 420.0 420.0 455.2
CDU cut point (F)
Diesel 610.0 610.0 610.0 630.0 610.0 610.0
VGO 1050.0 1050.0 1050.0 1050.0 950.0 1050.0
FCC unit 78.0 78.0 0.0 0.0 72.0 0.0
conversion (%)
Blending products (BPD)
PRG 15000 8880 15490 18855 15415 5000
RGG 15000 34670 5000 5000 10405 16280
ATK 25975 6740 37045 20470 6740 28405
LSDSL 29030 27810 26515 41180 18885 36620
LSFO 18980 22800 20085 18590 37820 18585
Refinery profit 2.1791 2.0252 2.1471 2.150 1.4319 2.2292
($MUS/D)

One of the main advantages of the nonlinear planning model, which can be inferred
from the previous discussion and results, is that it can provide an optimal operating
strategy for the refinery while at the same time meet product’s properties and production
rates. Quality give-away is also minimised hence resulting in large savings for the
petroleum refinery. This of course is in addition to the more accurate representation
of the refinery units.
 A nonlinear programming model for refinery planning 305

5 Conclusion

In this paper, an efficient nonlinear refinery-planning model has been presented.

The model represents the connections between processing units and intermediate streams
and optimises the operating variables of each individual unit. Different types of crudes
can be handled. The CDU fractions cut-point temperature and the operating variables
show the greatest effect on the final products flow rates and quality. The results from the
case studies show how the model can be utilised to minimise quality give-away.
The model proved to be computationally tractable and is able to integrate new processing
units or replace the existing units with new ones without major modifications. The future
work for this research will be to integrate hydrogen management strategies within
the refinery planning and study the overall effect on refinery operation. Another possible
extension is to consider uncertainties in raw materials availability, final product prices
and demands, and parameters in the refinery unit models.

Acronyms
CDU Crude Distillation Unit (Atmospheric and vacuum)
GP Gas Plant
NHT Naphtha Hydrotreater
DHT Distillate Hydrotreater
GOHT Gas Oil Hydrotreater
RDHT Residue Hydrotreater
CR Catalytic Reformer
HC Hydrocracker
FCC Fluid Catalytic Cracker
PRG Premium Gasoline
RGG Regular Gasoline
Kero Kerosene
ATK Aviation Turbine Kerosene
LSDSL Low Sulfur Diesel
LSFO Low Sulfur Fuel Oil
SUL Sulphur weight %
RON Research Octane No.
RVP Reid Vapor Pressure
OXG Oxygenate weight %
FLSH Flash Point Temp.
FRZ Freeze Point Temp.
CI Cetane Index
VISC Viscosity @ 210F
306 I. Alhajri et al.

References
Al-Enezi, G., Fawzi, N. and Elkamel, A. (1999) ‘Development of regression model to control
product yields and properties of the fluid catalytic cracking process’, Petroleum Science and
Technology, Vol. 17, No. 6–7, pp.535–552.
Allen, D.H. (1971) ‘Linear programming models for plant operations planning’, British Chemical
Engineering, Vol. 16, pp.685–691.
Baird, C. (1987) Petroleum Refining Process Correlations, HPI consultant.
Bodington, C.E. (1995) Planning, Scheduling, and Control Integration in the Process Industries,
McGraw-Hill, New York.
Brooke, A., Kendrick, D., Meeraus, A. and Ramesh R. (1998) GAMS–A User’s Guide: Tutorial by
Richard E. Rosenthal, GAMS Development Corporation, Washington DC, USA.
Drud, A.S. (1994) ‘CONOPT: a large scale GRG code’, Operations Research Society of America
Journal of Computing, Vol. 51, No. 11, pp.1271–1288.
Elkamel, A., Al-Ajmi, A. and Fahim, M. (1999) ‘Modeling the hydrocracking process using
artificial neural networks’, Petroleum Science and Technology, Vol. 17, Nos. 9–10,
pp.931–954.
Energy Information Administration (2006) World Oil Market and Oil Price Chronologies,
http://www.eia.doe.gov. accessed February.
Favennec, J.P. (2001) Refinery Operation and Management, Volume 5 of Petroleum Refining
Series, Editions Technip, Paris.
Gary, J.H. and Handwerk, G.E. (1994) Petroleum Refining: Technology and Economics, 3rd ed.,
Marcel Dekker, Inc. New York.
Li, W., Hui, C.W. and Li, A. (2005) Integrating CDU, FCC and product blending models into
refinery planning’, Computers and Chemical Engineering, Vol. 29, pp.2010–2028.
Maiti, S.N., Eberhardt, J., Kundu, S., Cadenhouse-Beaty, P.J. and Adams, D.J. (2001) ‘How to
efficiently plan a grassroots refinery’, Hydrocarbon Processing, Vol. 80, No. 6, pp.43–50.
Maples, R.E. (1993) Petroleum Refinery Process Economy, Penn Well Publishing Company,
Oklahoma, USA.
Moro, L.F.L., Zanin, A.C. and Pinto, J.M. (1998) ‘A planning model for refinery diesel
production’, Computers and Chemical Engineering, Vol. 22, pp.1039–1042.
Neiro, S.M.S. and Pinto, J.M. (2005) ‘Multiperiod optimization for production planning of
petroleum refineries’, Chemical Engineering Communications, Vol. 192, pp.62–88.
Pinto, J.M. and Moro, L.F.L. (2000) ‘A planning model for petroleum refineries’, Brazilian Journal
of Chemical Engineering, Vol. 17, Nos. 4–7, pp.575–586.
Riazi, M.R. (2005) Characterization and Properties of Petroleum Fractions, 1st ed., ASTM
International, PA, USA.
Symonds, G. (1956) ‘Linear programming solves gasoline refining and blending problems’,
Industrial and Engineering Chemistry, Vol. 48, No. 3, pp.394–401.
Watkins, R.N. (1979) Petroleum Refinery Distillation, 2nd ed., Gulf Pub. Co., Book Division,
Houston, USA.
Zhang, N. and Zhu, X.X. (2000) ‘A novel modelling and decomposition strategy for overall
refinery optimisation’, Computers and Chemical Engineering, Vol. 24, pp.1543–1548.
Zhang, J., Zhu, X.X. and Towler, G.P. (2001) ‘A simultaneous optimization strategy for overall
integration in refinery planning’, Industrial and Engineering Chemistry Research, Vol. 40,
pp.2640–2653.
 A nonlinear programming model for refinery planning 307

Nomenclature
I Set of units (i) in the refinery
J Set of units (j) that can send products to unit (i), J ∈ I
N Set of streams (s) of unit (i) can be sent to unit (j), N ∈ I
M Set of unit (m) can received stream (s) from unit (i), M ∈ I
B Set of final blending units (b), B ∈ I
E Set of unit (e) received external feed, E ∈ I
S Set of product streams (s) of unit (i)
P Set of properties (p) of stream (s)
PF Set of properties (p) of feed to unit (i)
X Set of operating variables of unit (i)
Indices
i, j, m, b For refinery unit
s, n For stream
P For property
X For operating variable
Parameters
Umaxi Maximum capacity of unit (i)
U
TE , TE
S
L
S
Upper and lower bounds of the end point temp. of stream (s) from
CDU
Cpi Cost price of product from unit (i)
Cxi Operating cost for unit (i)
Cfi Cost price of feed to unit (i)
Cpi Cost price of operating unit (i)
ak,p Coefficient for calculating the property (p) of stream (s)
Variables
Fi Volumetric flow rate of feed to unit (i), BPD
Vi,s Volumetric flow rate of stream (s) from unit (i), BPD
Wi,s Weight flow rate of stream (s) from unit (i), KLbPD
VSi,s,m Volumetric flow rate of stream (s) splitted from product Vi,s of unit
(i) received by unit (m), BPD
PVi,s,p Property (p) of stream (s) from unit (i)
FPi,p Property (p) of feed to unit (i)
XUi,u Operating variable x of unit I
TECDU,s End point (EP) cut temperature for stream (s) of (CDU) unit
Cuts Volume percent vapourised of stream (s) at the TEs
ys Weight or volume fraction of stream (s)