Sei sulla pagina 1di 16

ARTICLE

pubs.acs.org/IECR

Improvements in Petroleum Refinery Planning: 1. Formulation of


Process Models
Omar J. Guerra† and Galo A. C. Le Roux*
Department of Chemical Engineering, University of S~ao Paulo, Av. Prof. Luciano Gualberto trav.3, 380, CEP 05508-900,
S~ao Paulo-SP, Brazil

ABSTRACT: Optimization of the production planning is crucial for the economic success of a petroleum refinery. Nevertheless, it is
a difficult task, because of the large scale of the system and the complexity of the processes involved. Part 1 of this series of two papers
addresses the formulation of process models for petroleum refinery planning. First, a generic formulation of nonlinear refinery
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

planning model is adopted from the literature. Subsequently, the formulation of nonlinear empirical models for crude distillation
units (CDUs) and a fluid catalytic cracking (FCC) unit is addressed. These empirical models were successfully validated using
Downloaded via OPEN UNIV OF HONG KONG on January 26, 2019 at 14:20:14 (UTC).

rigorous process simulators. Finally, the results from model validation showed that the accuracy of model predictions is as good as
the current empirical process models reported in the literature, while the empirical process models proposed in this work overcome
the limitations of both linear and nonlinear empirical models for CDUs and FCC units previously developed by other authors. Part 2
[Ind. Eng. Chem. Res. 2011, DOI: 10.1021/ie200304v] involves the implementation of nonlinear empirical process models in the
petroleum refinery planning.

1. INTRODUCTION out in a FCC process simulator from PETROBRAS. Finally, the


For many years, production planning problems in petroleum predictions from the empirical models developed in this work
refineries have been addressed using the linear programming were compared with those obtained using empirical models from
(LP) technique.1 This technique is based on delta-base process the literature.
models (linearization of nonlinear process behavior at a parti-
cular set of operating conditions). Nevertheless, the linear
process models are not suitable for refinery process modeling, 2. REFINERY PLANNING
since refinery processes involve both physical operations (phase Being profitable in petroleum refining depends on using
separations, blending operations, etc.) and chemical operations systems and tools that enable the right people/functions to make
(cracking reactions, hydrotreating reactions, etc.) that are char- the right decisions at the right time.4 These decisions are taken at
acterized by their nonlinear nature. Because of this fact, the different hierarchical levels (advanced control, real-time optimi-
results (operating plans) from production planning models zation, production scheduling, production planning, and strate-
based on the linear programming technique can be unreliable. gic planning) and over different time horizons (seconds, minutes,
Furthermore, the recent advances in computer hardware and days, weeks, months, and years), using phenomenological or
optimization algorithms allow the implementation of nonlinear empirical process models. The model accuracy, the computa-
process models in the production planning models. Moreover, tional effort to solve it, and the profit scale for each decision level
Moro et al.2 quantify the impact of the adequate usage of refinery can be represented as shown in Figure 1. As can be seen in this
planning tools based on nonlinear process models in a potential figure, the profit scale is inversely proportional to the model
increase of ∼6 000 000 US$/yr in the profitability for a typical accuracy and the effort to solve it. In addition, there is an
refinery with operational capacity of 169.8 kbpd/day. Conse- interaction between each two adjacent hierarchical levels; for
quently, nowadays, different petroleum companies such as BP, example, the planning model provides the production plan for
ECOPETROL, ExxonMobil, and PETROBRAS have decided to the scheduling model, and this provides the production program
develop their own refinery planning tools (based on nonlinear and the feasibility of the production plan from the planning level.
process models), instead of using commercial tools (based on the According to Kreipl and Pinedo,5 the interaction between the
linear programming technique) such as the Process Industry planning and the scheduling levels may be intricate, because, in
Modeling System (PIMS) or the Refinery and Petrochemical the planning model, the feedstock supply and the final product
Modeling System (RPMS). demand and production are considered as average values for
In this paper, nonlinear empirical models for crude distillation the entire time horizon, whereas, in the scheduling model, the
units (CDUs) and fluid catalytic cracking (FCC) units are operation is considered to be a sequence of activities and the
proposed and validated. The empirical model for CDUs was
based on the empirical procedure proposed by Watkins,3 using Received: February 12, 2011
HYSYS (a rigorous process simulator) for model validation. In Accepted: July 22, 2011
addition, the empirical model for a FCC unit was based Revised: July 8, 2011
(formulation and validation) on rigorous simulations carried Published: July 22, 2011

r 2011 American Chemical Society 13403 dx.doi.org/10.1021/ie200303m | Ind. Eng. Chem. Res. 2011, 50, 13403–13418
Industrial & Engineering Chemistry Research ARTICLE

Figure 1. Hierarchical structure of the decisions in a petroleum refinery.

Figure 2. Process unit structure.2

feedstock supply and the final product demand and production The set USu0 ,u contains all streams from the unit u0 that can be
vary along the time horizon. sent as a feed to the unit u; similarly, the set SUs,u contains all
Recently, several works have been reported about the integra- units u0 that can receive (as a feed) the stream s from the unit u.
tion between planning and scheduling in order to reduce the dis- All streams existing in the entire refinery are characterized by the
crepancies in the results from these two models.68 Moreover, in set of properties p = {p1,p2,..,pm}, the relevant properties for the
order to provide accurate estimation of product yields and pro- feed to unit u are defined by the set PIu, while the relevant
perties in the planning model, it is necessary to ensure an ade- properties for each product stream s from this unit are defined by
quate modeling of the refinery process. Several works have been the set POu,s. In addition, the variables PAu,p and PCu,s,p denote
reported that are related to the implementation of nonlinear the property values for the feed and each product streams s in the
empirical process models in the refinery planning,2,9,10 because unit u, respectively. The set SOu contains all product streams
of the limitation of linear models to reproduce the refinery process from unit u, the yield and properties for the product stream
behavior under any particular set of operating conditions. Moro s∈SOu can be function of a set of operational variables Ou = {o1,
et al.2 presented a general framework for the formulation of o2,..,ok}. The value for each operational variable is denoted by the
nonlinear refinery planning models, which will be described below. variable Vo,u. The units that can be fed by external streams (i.e.,
2.1. Nonlinear Planning Model. According to Moro et al.,2 distillation units) are defined by the set Uf, as well as the units that
the topology of a refinery can be defined as a set of process units produce final products are defined by the set Up, these units can
interconnected by mixers and splitters; the structure of each be represented as mixers. For convenience, the feed streams to
process unit is described in Figure 2. In general, the streams from each unit u∈Uf are considered to be product streams from a
one or more units (u0 ∈Uu) that can be sent (at a flow rate Fu0 ,s,u) hypothetical unit r. Based on the sets and variables defined above,
as a feed to unit u are defined by the set su = {s1,s2,..,sn}. These the mathematical model for the refinery planning can be written
streams are mixed together to obtain the total feed FFu to the unit u. as follows:
13404 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418
Industrial & Engineering Chemistry Research ARTICLE

Table 1. Mixing Rules and Property Indicesa


property or property index notation addition base index correlation reference

cetane index CTI volume


motor octane numberb MON volume
research octane numberb RON volume
specific gravity SPG volume
API gravity API weight
Conradson carbon residue (wt %) CCR weight
sulfur (wt %) SUL weight
pour point index (pp = pour point (F)) PPI volume PPI = 316200 exp{12.5 ln[0.001(PP + 459.67)]} Li et al.11
smoke point index (smp = smoke point (mm)) SPI volume SPI = 362 + [2300/ln(SMP)] Li et al.11
viscosity index (ν = kinematic viscosity (cSt)) VII volume VII = log10(ν)/(3 + log10(ν)) Riazi12
a b
The mixing rules are a volumetric average, if the addition base is volume, or a weight average, if the addition base is weight. The mixing rule for these
properties is a simplification used in this work, usually these properties have a nonlinear mixing rule.

Objective function: unitary product prices, raw material costs, and operating costs
∑ ∑ ∑
respectively. These costs and prices are not necessarily constants.
max Profit ¼ Cpru 3 FF u  Crms 3 Fr, s, u For example, the product prices can be a function of a key
u ∈ Up u ∈ Uf s ∈ USr, u
property (in the case of gasoline, the octane number can be the
 ∑
u∈U
Copu 3 FF u ð1Þ key property) and the operating costs can be a function of
operational variables (for more details, see Part 2 of this work
subject to [Ind. Eng. Chem. Res. 2011, DOI: 10.1021/ie200304v]). Equa-
∑ ∑
tion 2 denotes the mass balance for the feed mixer in each process
FF u ¼ Fu0 , s, u "u∈U ð2Þ unit. Equation 3 represents the total feed properties as a function
u0 ∈ Uu s ∈ USu0 , u
of the properties and flow rate of each individual feed stream to
the unit. The function fu,p denotes a mixing rule for the property
PAu, p ¼ fu, p ðFu0 , s, u ju0 ∈ Uu ;s ∈ USu0 , u , PCu0 , s, p ju0 ∈ Uu ;s ∈ USu0 , u ;p ∈ PI u Þ p; this mixing rule can be on a volume basis (i.e., density, olefins
" u ∈ U, p ∈ PI u ð3Þ (vol%), etc.) or a weight basis (i.e., sulfur content (wt %),
nitrogen content (ppm), etc.). Nevertheless, many properties
involve highly nonlinear mixing rules (i.e., pour point, flash point,
FPu, s ¼ hu, s ðFF u , PAu, p jp ∈ PIu , Vo, u jo ∈ Ou Þ " u ∈ U, s ∈ SOu
viscosity, etc.); in this case, the linear blending indices (volume-
ð4Þ based or weight-based) can be used, the mixing rules and
property indices used in this paper are described in Table 1.
PCu, s, p ¼ gu, s, p ðPAu, p jp ∈ PI u , Vo, u jo ∈ Ou Þ " u ∈ U, s ∈ Su , p ∈ POu, s Equation 4 represents the flow rate of each product stream as a
function of operational variables Vo,u|o∈Ou and flow rate and
ð5Þ properties of total feed to the unit. Similarly, eq 5 represents the
properties of product streams PCu,p|p∈PIu as a function of

u0 ∈ SUs, u
Fu, s, u0 ¼ FPu, s " u ∈ U, s ∈ SOu ð6Þ operational variables Vo,u|o∈Ou and properties of the total feed
to the unit. Equation 6 denotes the mass balance for each product
stream s from each unit. Inequality 7 denotes the lower bounds
FF u g Demandu " u ∈ Up ð7Þ on product demands for blending units (u∈Up); it will be
assumed that product demand is flexible in the sense that it is

u ∈ Uf
Fr, s, u e Supply s " s ∈ USr, u ð8Þ given by a hard lower bound and a free upper bound. Similarly,
inequality 8 denotes the upper bounds on raw material supply.
Inequality 9 represents the operational capacity for every unit in
the refinery. Inequality 10 denotes the specifications for the
u e FF u e FF u
FF lo "u∈U ð9Þ
up
final product streams (u∈Up), as well as the possible feed
specification for the process units (uˇUp). In addition, inequal-
u, p e PAu, p e PAu, p
PAlo " u ∈ U, p ∈ PI u ð10Þ
up
ity 11 denotes the lower and upper bounds for the operational
variables in each process unit. These limits generally are subject
Vo,lou e Vo, u e Vo,upu " u ∈ U, o ∈ Ou ð11Þ to design conditions. Finally, expression 12 denotes the domain
of the variables used in the formulation of the refinery
FF u , Fu0 , s, u , FPu, s ∈ Rþ ; PAu, p , PCu, s, p , Vo, u ∈ R ð12Þ planning model.
The mathematical model for the refinery planning described
above results in a nonlinear planning model and can be solved
Equation 1 defines the objective function (maximize the total using nonlinear programming (NLP) optimization algorithms
profit (US $/day)) for the refinery planning model in terms of such as sequential linear programming (SLP), generalized re-
sales revenue (blending units, u∈Up), raw material purchase duced gradient (GRG), sequential quadratic programming
costs (units that can process external streams, u∈Uf), and (SQP), and interior point (IP). These algorithms have been
operating costs. The terms Cpru, Crms, and Copu denote the implemented in some NLP solvers, such as CONOPT, IPOPT,
13405 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418
Industrial & Engineering Chemistry Research ARTICLE

Figure 3. Simplified distillation processes.

KNITRO, MINOS, and SNOPT, which are available in many adjacent fractions is optimized using a pseudo-cut called a swing
modeling systems (AIMS, AMPL, GAMS, and MINOPT). cut (a virtual cut between each two adjacent fractions) with
constant properties. The size of each swing cut is defined as a
3. CRUDE DISTILLATION UNITS certain volume percentage or weight percentage of the overall
crude fed to the CDU (i.e., 5 wt% for the swing cut naphtha/
Crude distillation units (CDUs) are the most important
kerosene, according to Zhang et al.13). Optionally, the size of
process units in a refinery, because they perform the separation
of crude oil in many fractions (naphtha, kerosene, diesel oil, each swing cut can be defined as a certain TBP range (i.e., 50 F,
atmospheric residue, vacuum gas oils, etc.) that are processed in according to Trierwiler14). In the case that more than one crude
the downstream units (FCC units, delayed coking unit, etc.) or oil can be fed to the CDU, the flow rate of each swing cut from all
sent to the blending system to form final products. Conse- crude oils are pooled and the resultant stream is distributed
quently, the overall economical performance of a refinery is between the corresponding two adjacent fractions. The major
directly impacted by the operation of the CDUs. limitation of the swing cut model is that the properties for each
The crude distillation process occurs in two sections: in the swing cut are considered constants across its TBP range, while
first one (denoted as atmospheric distillation), the operating pres- the property distribution across the TBP ranges are known to be
sure is set close to atmospheric pressure. In the second section highly nonlinear. Consequently, the swing cut model fails to
(denoted as vacuum distillation), the operating pressure is set to represent the highly nonlinear behavior of distillation processes.
vacuum pressure (generally in the range of 50100 mmHg). Prior The adherent recursion model addresses this limitation, using
to the distillation processes, the crude oil is desalted using water an iterative procedure between a LP planning model and a rig-
(to prevent corrosion problems) and heated to provide the ade- orous process simulator (HYSYS, PRO/II, etc.). In this proce-
quate conditions for the separation. In Figure 3, a simplified dure, the LP planning model is used to optimize the TBP cut
distillation process is presented. The yields and properties of the points and the solution at each iteration (new cut points) is used
different fractions from the distillation process depend on the type by the rigorous process simulator to provide updated yields and
of crude processed (heavy, medium, or light) and the operating properties for the CDU fractions. The main limitation of the
conditions (heater outlet temperature, reflux ratio, etc.). adherent recursion is the long computational time required for
3.1. Overview of Empirical Models for Crude Distillation the convergence of the iterative procedure, because of the use of a
Units. The empirical models for CDUs reported in the literature rigorous process simulator.
include the swing cut model,13 the adherent recursion model,14 Finally, in the WTR model, the CDU operating modes are
and the Weight Transfer Ratio (WTR) model.15 The swing cut used to determine the maximum and minimum yield of each
model is the most commonly used of these models. Normally, fraction over the TBP curve of the crude oil fed to the CDU;
the swing cut model is implemented in commercial planning these values are fixed as upper and lower bounds for the yield of
tools such as RPMS and PIMS, basically due to its simplicity. On each fraction from the CDU. In addition, the properties of CDU
the other hand, the swing cut model is more inaccurate than the fractions are expressed as a nonlinear function of their respective
adherent recursion and WTR models. In the swing-cut model, mid-volume transfer ratio. The TBP cut points are optimized
the TBP (true boiling point curve) cut point between each two using the yield of CDU fractions as decision variables constrained
13406 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418
Industrial & Engineering Chemistry Research ARTICLE

Figure 4. Typical TBP distillation curve. Figure 5. Typical sulfur distribution curve.

by the mass balance and the maximum and minimum yields The definition of the TBP cut point between adjacent fractions
obtained from the operating modes. Therefore, this model can re- is illustrated in Figure 6. In this figure, the temperature T100 L
present the highly nonlinear behavior of distillation processes. represents the final TBP boiling point for the light fraction
Nevertheless, as in the petroleum industry, it is very usual that the (EBPL) and the initial TBP boiling point for the heavy fraction
CDU is able to process more than one crude oil; the implementa- (IBPH) is represented by the temperature T0H. As can be seen in
tion of the WTR model requires additional tedious procedures to Figure 6, the TBP cut points are defined as the average
estimate the TBP curve of the crude oil mixture fed to the CDU. temperature between the EBPL and IBPH (Tcut = 0.5(T100 L 
3.2. Empirical Model for CDUs. The necessary characteriza- T0H)). In addition, the degree of separation between two
tion information of each crude oil must be available in the adjacent fractions is given by the difference between EBPL and
refinery (i.e., from laboratory databases) in order to developed IBPH (T100 L  T0H). This degree of separation depends upon
a meaningful empirical model for CDUs. There are two types of the efficiency of the distillation column. The detailed procedure
characterization information normally used for crude oil evalua- for the estimation of the TBP cut points is presented in Figure 7.
tions: the crude oil distillation curve (TBP) and the crude oil As shown in this figure, the TBP cut points can be estimated from
property distribution curves. ASTM D86 100% (the temperature corresponding to the 100
The TBP curve is the relationship between the distillate tem- vol % on the ASTM D86 distillation curve) product specifica-
perature and the cumulative volumetric fraction distilled in a tions and the degree of separation given by the Gap (595)
standard distillation column. The standard test method for the ASTM (the temperature difference between the ASTM 5 vol %
experimental estimation of the TBP is ASTM D2892 (at atmo- boiling point of the heavier product and the ASTM 95 vol %
spheric pressure) for fractions with boiling points below 400 C boiling point of the adjacent lighter product).
and the ASTM D1160 (at reduced or vacuum pressure) for The product yields and properties are strongly affected by
fractions with boiling points above 400 C. For distillate pro- changes in TBP cut points. Furthermore, they are commonly
ducts, standard method ASTM D86 is generally used. Distillation used in the CDU operation to control the adequate separation
curves obtained by the ASTM D1160 and ASTM D86 standard efficiency in the distillation column, as well as the product
test methods can be converted into TBP curves using the method specifications. For this reason, the TBP cut points are chosen
described by Riazi.12 A typical TBP curve is presented in Figure 4. as decision variables in the proposed approach.
On the other hand, the properties (API density, sulfur (wt %), The ASTM D86 specifications are defined for each refinery,
etc.) of distillate products can be expressed using the property and the Gap (595) can be set by heuristics or estimated using a
distribution curves, which represent the relationship between the rigorous process simulator (HYSYS, PRO/II, etc.). In that case,
crude oil properties and the mid-volume transfer ratio (Mid-Vol many scenarios must be defined by varying the operating con-
(%)) of the fraction distilled.3 A typical sulfur (wt %) distribution ditions and the feedstock characterization in order to evaluate the
curve is presented in Figure 5. variability of each degree of separation; the Gap (595) ASTM
The TBP curve and the property distribution curves for each between each two adjacent fractions should be reported as the
crude oil (available in the refinery) are the necessary character- average value for all of the scenarios. Usual values (recommended
ization information for the empirical model developed in this by Watkins3) for the ASTM D86 boiling point range (IBP and
work. In many cases, the distribution curve for a given property EBP) of CDU fractions and the Gap (595) ASTM are
can be unavailable in the refinery. In that case, the distribution presented in Tables 2 and 3, respectively. In this work, ranges
curve can be predicted from the TBP curve and the distribution of 260275 F and 650700 F will be considered for the EBP
curve for other property (density, viscosity, molecular weight, of gross overhead and heavy distillate, respectively; in other case,
etc.) using empirical correlations.12 the values from Table 2 will be used.
In addition to the crude oil characterization information, the set The volumetric yield of each CDU fraction can be estimated as
of decision variables is another important aspect to be defined in a function of the TBP cut points. For example, the volumetric
the formulation of the empirical model for CDU. This set should yield (vol %) of heavy naphtha for a given crude oil is estimated as
be related to the operating variables that significantly affect the the difference between the cumulative volume percent corre-
process behavior for a given feedstock characterization. The TBP sponding to the two adjacent TBP cut points (heavy naphtha/
cut points between adjacent fractions are probably the operating light distillate and gross overhead/heavy naphtha TBP cut
variables that have the greatest impact on the CDU operation. points). In addition, the Mid-Vol (%) value of each fraction is
13407 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418
Industrial & Engineering Chemistry Research ARTICLE

Figure 6. Definition of TBP cut point between adjacent fractions.3

Figure 7. Procedure for the estimation of TBP cut point between adjacent fractions.3

estimated using its volumetric yield and the actual cumulative while the relevant properties for each product stream are defined
volume percentage. by the set POCDU,s⊆POCDU. The variables PCCDU,s,p and PDs0 ,s,p
3.2.1. Mathematical Model. The mathematical formulation of represent the property values of the product streams and the
the empirical model for the CDUs is based on the process unit contribution of each crude oil to the property values of each
structure presented in Figure 2. It is considered that the refinery product stream from the CDU, respectively. The units that can
has only one CDU, and this is the only unit that processes crude process product streams from the CDU are defined by the set
oil (Uf = {CDU}). Nevertheless, the model can be easily SUs,CDU. The operating variables in the CDU are defined by the
implemented if there is more than one CDU in the refinery. set of TBP cut points between adjacent fractions OCDU =
The model was formulated on a volumetric basis (volumetric {Ts1,s2,Ts2,s3,...,Tsn1sn}. In addition, the variable VTsi,si+1 ,CDU de-
balances are very common in the petroleum industry), then all of notes the value of each operating variable in the CDU.
the flow rates are given in volumetric units. The set USr,CDU = The crude oils are characterized by the TBP curve (TBPs0 )
{s01 ,s0 2,...,s0 k} defines the crude oils (at a flow rate Fr,s0 ,CDU) that and the property distribution curves (PROp,s,). The variable
can be fed into the CDU, and the total feed flow rate to the CDU VDs0 ,Tsi ,si+1 represents the volume percent vaporized of crude oil s0
for each TBP cut point and the variable Mid_Vols,s0 represents the
is represented by the variable FFCDU. The product streams from
mid-volume transfer ratio for product streams from the CDU.
this unit are defined by the set SOCDU = {s1,s2,...,sn}. In addition,
Finally, based on the sets and variables described above, the
the variable FPCDU,s denotes the total flow rate of each product
mathematical model for the CDUs can be written as follows:
stream from the CDU, and the contribution of each crude oil to
the total flow rate of each product stream is represented by the VDs0 , Tsi , siþ1 ¼ TBPs0 ðVTsi , siþ1 , CDU Þ " s0 ∈ USr, CDU , Tsi , siþ1 ∈ OCDU
variable FDs,s0 . In general, the product streams from the CDU can
be characterized by the set of properties POCDU = {p1,p2,...,pm}, ð13Þ

13408 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418


Industrial & Engineering Chemistry Research ARTICLE

Table 2. Suggested ASTM Boiling Ranges for CDU Fractionsa

Operating Mode

product max naphtha max light distillate max heavy distillate

gross overhead (GO) 250275 F (EBP) 250275 F (EBP) 250275 F (EBP)


heavy naphtha (HN) 400 F (EBP) 325 F (EBP) 325 F (EBP)
light distillate (LD) 375600 F (IBPEBP) 300600 F (IBPEBP) 300550 F (IBPEBP)
heavy distillate (HD) 575675 F (IBPEBP) 575675 F (IBPEBP) 525675 F (IBPEBP)
a
Data taken from ref 3.

Table 3. Suggested Gap (595) ASTM for Adjacent


Fractionsa

u0 ∈ SU s, CDU
FCDU, s, u0 ¼ FPCDU, s " s ∈ SOCDU ð23Þ

adjacent fractions Gap (595) ASTM (F) up


VTlosi , s e VTsi , siþ1 , CDU e VTsi , siþ1 , CDU " Tsi , siþ1 ∈ OCDU
iþ1 , CDU
gross overhead/heavy naphtha (GO/HN) 2030
heavy naphtha/light distillate (HN/LD) 2550 ð24Þ
light distillate/heavy distillate (LD/HD) 010
" s0 ∈ USr, CDU
up
heavy distillate/bottom residue (HD/BR) 010 Fr, s0 , CDU e Fr, s0 , CDU ð25Þ
a
Data taken from ref 3.
FF CDU ¼ ∑
s0 ∈ USr, CDU
Fr, s0 , CDU ð26Þ
 
1
FDsi , s0 ¼ Fr, s0 , CDU ðVDs0 , Tsi , siþ1  VDs0 , Tsi1 , si Þ FF loCDU e FF CDU e FF CDU
up
ð27Þ
100
"si jTsi , siþ1 ∈ fTs2 , s3 , Ts4 , s5 , :::, Tsn2 , sn1 g, s0 ∈ USr, CDU
FF CDU , Fr, s0 , CDU , VDs0 , Tsi , siþ1 , FDsi , s0 , Midvolsi , s0 , FCDU, s, u0 ,
ð14Þ
FPCDU, s ∈ Rþ ; PDs0 , s, p , PCCDU, s, p , Vo, CDU ∈ R ð28Þ
 
1 Equation 13 defines the volume percent vaporized (in units of
FDs1 , s0 ¼ Fr, s0 , CDU VDs0 , Ts1 , s2 " s0 ∈ USr, CDU
100 vol %) of each crude oil as a function of the TBP cut points (in
  this work, the TBP curves are represented using a third-order
ð15Þ
1 polynomial function). The contribution of each crude oil to the
FDsn , s0 ¼ Fr, s0 , CDU ð100  VDs0 , Tsn1 , sn Þ " s0 ∈ USr, CDU total flow rate of each product stream is described by eqs 14, 15,
100
and 16, while the total flow rate for these products is obtained
ð16Þ
using eq 17. In addition, the mid-volume transfer ratio for
product streams can be estimated according to eqs 18, 19, and
FPCDU, s ¼ ∑
s0 ∈ USr, CDU
FDs, s0 " s ∈ SOCDU ð17Þ 20. The contribution of each crude oil to the property values of
each product stream is expressed by eq 21; in this work, the
property distribution curves were represented using a fourth (or
Mid _ Volsi , s0 ¼ VDs0 , Tsi1 , si þ 0:5ðVDs0 , Tsi , siþ1  VDs0 , Tsi1 , si Þ less)-order polynomial function. Furthermore, the properties of
ð18Þ product streams are estimated using eq 22 if the mixing rule of
"si jTsi , siþ1 ∈ fTs2 , s3 , Ts3 , s4 ,..., Tsn2 , sn1 g, s0 ∈ USr, CDU the property is based on volume. In the case of properties with a
mixing rule based on weight, the flow rate in eq 22 should be
Mid_Vols1 , s0 ¼ 0:5VDs0 , Ts1 , s2 " s0 ∈ USr, CDU ð19Þ given in mass units (the property indices shown in Table 1 can be
used for properties with highly nonlinear mixing rules). Equation
23 denotes the volumetric balance at the product splitters.
Mid_Volsn , s0 ¼ 100  0:5VDs0 , Tsn1 , sn " s0 ∈ USr, CDU Inequalities 24 and 25 denote the constraints for operating
ð20Þ variables and crude oil supply, respectively. Equation 26 repre-
sents the volumetric balance at the feed mixer (crude oil mixer)
PDs0 , s, p of the CDU. The operating capacity is defined by inequality 27.
Finally, the expression described by eq 28 denotes the domain of
¼ PROp, s0 ðMid_Vols, s0 Þ " s0 ∈ USr, CDU , s ∈ SOCDU , p ∈ POCDU, s the variables used in the formulation of the empirical model
for CDUs.
ð21Þ
3.2.2. Model Validation Using HYSYS. The empirical model
for the CDUs was validated using rigorous process simulations
PCCDU, s, p (based on TBP data and pseudo-components) carried out in
Aspen HYSYS v7.0, using PengRobinson as the fluid package.

1
¼ PDs0 , s, p FDs, s0 " s ∈ SOCDU , p ∈ POCDU, s
FPCDU, s s0 ∈ USr, CDU
Three case studies were evaluated using two different crude oils:
Tia Juana Light and Arabian Heavy. The configuration of each
ð22Þ case study is presented in Table 4. The main products from the
13409 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418
Industrial & Engineering Chemistry Research ARTICLE

Table 4. Case Studies for CDU Model Validation predictions using the empirical model was e4.10 vol %. This
absolute error can be reduced to be e0.51 vol % via the addition
Feed Flow Rate (kbpd) of bias factors to the empirical model predictions, as described in
eq 29. These bias factors were calculated as an average of the
crude oil Case A Case B Case C
empirical model errors, with respect to the rigorous simulation.
Tia Juana Light (API = 31.6) 60 70 80
Arabian Heavy (API = 27.4) 40 30 20 FPCDU, s ¼ ∑
s0∈ USr, CDU
FDs, s0 þ BiasðsÞ
total 100 100 100

CDU Feed Properties


 FF CDU " s ∈ SOCDU ð29Þ

Case A Case B Case C


Equation 29 is a modification of eq 17; the sum of the bias
API density 29.65 30.06 30.48 factors should be equal to zero, because of the volumetric balance
molecular weight 234.0 kg/kmol 230.2 kg/kmol 226.4 kg/kmol in the product splitters. In the work of Li et al.,15 the yield
predictions from the WTR model were compared (using three
Table 5. Product Specifications case studies) with rigorous simulation carried out in Aspen Plus
version 11.1; the absolute error in the yield predictions was
Specification determined to be less than or equal to 1.3, 3.2, 3.07, 2.67, and
3.06 wt % for the gross overhead (GO), heavy naphtha (HN), light
product ASTM D86, 100% Gap (595) ASTM distillate (LD), heavy distillate (HD), and bottom residue (BR),
gross overhead 275 F (135.0 C)
respectively. Finally, based on the results from model validation, it
can be concluded that the empirical model for CDU developed in
heavy naphtha 325 F (162.8 C) 25 F
this work is adequate for the simulation of the CDUs.
light distillate 600 F (315.6 C) 35 F
heavy distillate 700 F (371.1 C) 10 F
4. FLUID CATALYTIC CRACKING (FCC) UNIT
bottom residue 5 F
Fluid catalytic cracking (FCC) is the most important con-
version process in a refinery. In this process, heavy fractions
CDU are: gross overhead, heavy naphtha, light distillate, heavy (atmospheric gas oil, vacuum gas oil, cocker gas oil, etc.) are
distillate, and bottom residue. The specifications for CDU pro- converted into light and more-valuable fractions (ethane, ethy-
ducts and the degree of separation between adjacent fractions are lene, propylene, butylenes, iso-butane, n-butane, naphtha, etc.).
presented in Table 5. The data shown in Table 5 should be FCC naphtha becomes a gasoline blending stock. According to
interpreted as follows: the Gap (595) ASTM between gross Robinson,16 the FCC units produce more than half of the world’s
overhead and heavy naphtha is equal to 25 F. gasoline. Light olefins (particularly propylene and butylenes) and
The configuration of the CDU in HYSYS includes a preflash iso-butane are alkylated (in the alkylation unit) to produce high-
column, a crude oil heater, and a distillation column. The pre- octane hydrocarbons for the gasoline pool. In addition, ethane
heated crude oils are mixed and fed to the preflash column at and ethylene are used as petrochemical feedstock to produce a
450 F and 75 psia. In the preflash column, light components are variety of products.
removed from the crude oil (mixed crude oil) in order to reduce The FCC units consist of a riser reactor, a regenerator, and a
the energy consumption in the crude oil heater. The heavy product separation system (see Figure 9). In the riser, the fresh
product (bottom product) from the preflash column is fed to the feed is mixed with the regenerated catalyst to provide the nec-
crude oil heater, where it is heated to the appropriate column essary conditions for the cracking reactions. In the regenerator,
feed temperature. According to Watkins,3 the column feed the combustion air provides the necessary oxygen for the catalyst
temperature should be in the range of 650700 F in order to regeneration by burning of the coke in the spent catalyst from the
avoid thermal cracking. In this work, the column feed tempera- reactor. In addition, the cracked products from the reactor are
ture was set to 700 F. In addition, the distillation column obtained separately in the fractionator. The FCC feed composi-
includes a main column with 29 theoretical stages, three pump- tion has a direct impact in the process behavior. For example, the
arounds, three side-strippers, a condenser, and a reboiler (see reactivity increases according to paraffins > naphthenes > aro-
Figure 8). The distillation column has 13 degrees of freedom (see matics. Therefore, paraffinic feeds produce more naphtha (with
Table 6), while the empirical model developed in this work has less octane number) than naphthenic and aromatic feeds for the
only 8 degrees of freedom (4 ASTM D86 100% and 4 Gap same set of operating conditions. Olefins are not the preferred
(595) ASTM). This is because the empirical models are more feed to an FCC unit, because they often polymerize to form
simplified than the rigorous process models. For example, the undesirable products, such as decanted oil and coke. The typical
energy balance is not included in the empirical model for CDUs olefin content of FCC feed is <5 wt %.17
developed in this work. 4.1. Overview of Empirical Models for FCC Units. The
The CDU was simulated for all of the case studies using both empirical approaches that have been reported in the literature for
the rigorous process simulator model (the HYSYS model) and modeling FCC units include the delta-base model (commonly
the empirical model developed in this work. In the empirical implemented in RPMS and PIMS), the UltraCat model (from
model, the TBP cut points were estimated using the procedure Wollaston et al.18), and the procedure described by Gary and
described in Figure 7, using the specifications presented in Handwerk.19 The limitations of delta-base model were pre-
Table 5. The comparison of product yields is shown in Table 7. viously described in section 1 of this paper. Contrary to the delta-
As can be seen in this table, the absolute error in the yield base model, the UltraCat model is based on highly nonlinear
13410 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418
Industrial & Engineering Chemistry Research ARTICLE

Figure 8. CDU simulation model in HYSYS.

Table 6. Distillation Column Specifications numerical approximations. In addition, the model does not con-
sider the effect of the reactor temperature and the catalyst-to-oil
specification number of specifications
ratio (two of the most important operating variables in the FCC
gap (595) ASTM (F) 4 (GO-HN, HN-LD, LD-HD, and HD-BR) process) on product yields and properties, which can be a limitation.
ASTM D86, 100% (F) 4 (GO, HN, LD, and HD) 4.2. Empirical Model for a FCC Unit. In the FCC reactor, a
reflux ratio 1 (condenser) large number of chemical reactions involving a large number of
pump-around flow rate 1 (PA_1) hydrocarbon compounds take place. Therefore, the modeling of
pump-around heat duty 2 (PA_2 and PA_3) FCC process results in a large-scale modeling problem. Rigorous
off-gas flow rate 1 (condenser)
FCC process simulators such as Aspen FCC (AspenTech) and
FCC-SIM (KBC Advanced Technologies) are based on the
total degrees of freedom 13
lumping technique for feedstock and products characterization,
in order to reduce the inherent problem size. The number of
empirical correlations for the second-order conversion and the lumps used for the modeling of FCC process determines the
coke yield. These two empirical correlations (reactor model) are model structure and applicability. For example, in the 10-lumps
integrated with two equations that correspond to the heat model,20 the reaction rate constants can be considered to be
balance between the reactor and the regenerator; the yield and independent of the feed composition, which is very important for
properties for the remaining products are estimated using the simulation of industrial FCC units, because these units can
simplified empirical correlations. This model was successfully process both light and heavy feeds. Moreover, when more than
validated using both pilot- and industrial-plant data. The main 20 lumps are used, all reaction rates can be considered to be
limitation of the UltraCat model is that it is highly nonlinear and pseudo-first-order21 and the FCC model becomes mostly linear.
its integration into the entire refinery planning model can lead to However, the higher the number of lumps, the higher is the
excessive computational cost or difficulties in achieving a local number of reaction rate constants to be estimated. For the
optimal solution. regenerator, a rigorous coke burning kinetics (approximately
The procedure described by Gary and Handwerk19 is a graph- six reactions) is adopted. Because of the detailed modeling,
ical method where product yields and properties are estimated as rigorous FCC process simulators are suitable for the simulation
a function of the conversion (the operational variable) and two of industrial FCC units. Typically, they are used for catalyst
feed characterization parameters (API gravity and Watson char- selection, feed selection, delta-base model generation, and other
acterization factor). This graphical method can be impractical for applications.
refinery planning, because it is based on charts and figures The empirical model for a typical FCC unit developed in this
that should be converted into mathematical expressions using work consists of a set of empirical correlations based on data from
13411 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418
Industrial & Engineering Chemistry Research ARTICLE

Table 7. Comparison of Product Yields

Product Yield (vol %)

HYSYS Simulation Empirical Model Maximum Absolute Error (vol %)

producta Case A Case B Case C Case A Case B Case C without bias with bias bias (vol %)

GO 14.22 14.19 14.17 11.84 12.32 12.80 2.39 0.51 1.87


HN 5.32 5.38 5.46 4.91 4.90 4.89 0.57 0.08 0.49
LD 23.83 24.17 24.51 20.25 20.33 20.41 4.10 0.26 3.84
HD 10.59 10.73 10.86 14.41 14.54 14.67 3.82 0.01 3.81
BR 46.04 45.52 45.01 48.59 47.91 47.23 2.54 0.17 2.39
a
GO = gross overhead; NH = heavy naphtha; LD = light distillate; HD = heavy distillate; and BR = bottom residue.

Figure 9. Simplified fluid catalytic cracking (FCC) unit.

a PETROBRAS FCC process simulator (a rigorous process process unit structure presented in Figure 2. The product yields
simulator). The data consist of a series of simulations, using a set and properties were expressed as polynomial functions of
of independent variables (operating variables and feed proper- three or less independent variables. The FCC unit can be fed
ties). This set of independent variables was defined as the (at a flow rate Fu0 ,s00 ,FCC) by a set of intermediate streams
operating variables and feed properties that more affect the USu0 ,FCC = {s001 ,s002 ,...,s00w} from upstream units u0 ∈UFCC (UFCC =
FCC process behavior. The independent variables and their {CDU,...}). The total FCC feed can be characterized by a set of
respective operating limits (design specifications and operating properties PIFCC = {SPG,API,SUL,CCR}, where SPG, API, SUL,
constraints) are presented in Table 8. A total of 193 simulations and CCR denotes the specific gravity, API gravity, sulfur content
was used: 135 (70% of total simulations) were used for (in units of wt %), and Conradson carbon residue (in untis of
parameter estimation, and 58 (30% of total simulations) were wt %), respectively. The operating variables in the FCC unit are
used for model validation. The numerical values for all inde- defined by the set OFCC = {Trxn,CO}, where Trxn and CO
pendent variables in each simulation were defined according represents the reactor temperature and the catalyst-to-oil ratio,
to eq 30: in that order. The numeric values for the reactor temperature and
  catalyst-to-oil ratio are represented by the variables VTrxn,FCC and
up
xi, j ¼ xloi þ randi, j xi  xloi VCO,FCC, respectively. The product streams from the FCC unit
are defined by the set SOFCC = {C3T,C4T,NAP,LCO,DO,COK},
" i ∈ fAPI ,..., Trxn g, j ∈ f1, 2 ,..., 193g ð30Þ where C3T, C4T, NAP, LCO, DO, and COK denotes total C3s,
where xup lo
i is the upper bound for independent variable i, xi the total C4s, naphtha, light cycle oil, decanted oil, and coke respec-
lower bound for independent variable i, randi,j a random number tively. The relevant properties for naphtha, light cycle oil, and
between 0 and 1 for independent variable i in simulation j, and xi,j decanted oil are defined by the sets POFCC,NAP = {SPG,API,SUL,
the value for variable i in simulation j. RON,MON}, POFCC,LCO = {SPG,API,SUL,CTI,PPI,VII} and
4.2.1. Mathematical Model. Similar to the formulation of the POFCC,DO = {SPG,API,SUL,VII}, correspondingly. The indices
empirical model for CDUs, the formulation (on a volumetric RON, MON, CTI, PPI, and VII denote research octane number,
basis) of the empirical model for the FCC unit is based on the motor octane number, cetane index, pour point index, and viscosity
13412 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418
Industrial & Engineering Chemistry Research ARTICLE

index, respectively. The property indices are estimated according to FCLCO ¼ βLCO, 1 þ βLCO, 2 X 0 þ βLCO, 3 X 0 X 0 ð36Þ
Table 1. In addition, there are three auxiliary variables: weight
conversion (X), second-order conversion (X0 ), and weight hourly FCDO ¼ βDO, 1 þ βDO, 2 X 0 þ βDO, 3 X 0 X 0 ð37Þ
space velocity (WHSV).
Finally, based on the sets and variables described above, the FCCOK ¼ βCOK, 1 þ βCOK, 2 PAFCC, CCR
mathematical model for the FCC unit can be written as follows:
þ βCOK, 3 VCO, FCC þ βCOK, 4 WHSV
FF FCC ¼ ∑ ∑
u0 ∈ UFCC s00 ∈ USu, , FCC
Fu0 , s00 , FCC ð31Þ
þ βCOK, 5 PAFCC, CCR VCO, FCC

þ βCOK, 6 PAFCC, CCR WHSV


∑ ∑
1
PAFCC, p ¼ PCu0 , s00 , p Fu0 , s00 , FCC " p ∈ PIFCC
FFFCC u0 ∈ UFCC s00 ∈ USu0 , FCC
þ βCOK, 7 VCO, FCC WHSV ð38Þ
ð32Þ

X ¼ R1 þ R2 PAFCC, API þ R3 VCO, FCC


FCC3T ¼ βC3T, 1 þ βC3T, 2 X 0 þ βC3T, 3 X 0 X 0 ð33Þ
þ R4 VTrxn , FCC þ R5 PAFCC, API VCO, FCC
þ R6 PAFCC, API VTrxn , FCC þ R7 VCO, FCC VTrxn , FCC ð39Þ
FCC4T ¼ βC4T, 1 þ βC4T, 2 X 0 þ βC4T, 3 X 0 X 0 ð34Þ
X
X0 ¼ ð40Þ
FCNAP ¼ βNAP, 1 þ βNAP, 2 PAFCC, API 100  X

þ βNAP, 3 VCO, FCC þ βNAP, 4 X 0 kPAFCC, SPG FF FCC


WHSV ¼ ð41Þ
CatInv
þ βNAP, 5 PAFCC, API VCO, FCC

þ βNAP, 6 PAFCC, API X 0 þ βNAP, 7 VCO, FCC X 0 ð35Þ PCFCC, NAP, API ¼ βNAP, 8 þ βNAP, 9 PAFCC, API

Table 8. Range of Feed Properties and Operating Variables þ βNAP, 10 X ð42Þ


lower upper
variable bound bound PCFCC, NAP, SUL ¼ βNAP, 11 þ βNAP, 12 PAFCC, SUL
API gravity 19 27 þ βNAP, 13 X ð43Þ
total sulfur (wt %) 0.5 3.1
Conradson carbon 2.0 6.0
residue, CCR (wt %)
PCFCC, NAP, RON ¼ βNAP, 14 þ βNAP, 15 PAFCC, API
catalyst-to-oil ratio, C/O 3.85 6.8
þ βNAP, 16 X ð44Þ
weight hourly space 2.8 5.7
velocity, WHSV (h1)
reaction temperature, Trxn 956 F (513.3 C) 1000 F (537.8 C) PCFCC, NAP, MON ¼ βNAP, 17 þ βNAP, 18 PCFCC, NAP, RON ð45Þ

Figure 10. Parity plots for C3T and C4T yield prediction.

13413 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418


Industrial & Engineering Chemistry Research ARTICLE

Figure 11. Parity plots for naphtha and LCO yield prediction.

Figure 12. Parity plots for DO and coke yield prediction.

PCFCC, LCO, API ¼ βLCO, 4 þ βLCO, 5 PAFCC, API PCFCC, DO, API ¼ βDO, 4 þ βDO, 5 PAFCC, API

þ βLCO, 6 X ð46Þ þ βDO, 6 X ð51Þ

PCFCC, LCO, SUL ¼ βLCO, 7 þ βLCO, 8 PAFCC, SUL PCFCC, DO, SUL ¼ βDO, 7 þ βDO, 8 PAFCC, SUL

þ βLCO, 9 X ð47Þ þ βDO, 9 X ð52Þ

PCFCC, LCO, CTI ¼ βLCO, 10 þ βLCO, 11 PAFCC, API PCFCC, DO, VII ¼ βDO, 10 þ βDO, 11 PAFCC, API

þ βLCO, 12 X ð48Þ þ βDO, 12 X ð53Þ

  !
PCFCC, LCO, PPI ¼ βLCO, 13 þ βLCO, 14 PAFCC, API FPFCC, s ¼ FCs
1 PAFCC, SPG
FF FCC " s ∈ SOFCC
100 PCFCC, s, SPG
þ βLCO, 15 X ð49Þ
ð54Þ
PCFCC, LCO, VII ¼ βLCO, 16 þ βLCO, 17 PAFCC, API

þ βLCO, 18 X ð50Þ

u0 ∈ SU s, FCC
FFCC, s, u0 ¼ FPFCC, s " s ∈ SOFCC ð55Þ

13414 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418


Industrial & Engineering Chemistry Research ARTICLE

up
FF loFCC e FF FCC e FF FCC ð56Þ the formulation of the empirical model for the FCC unit. The
parameter in the model are defined by βs,j (parameters for product
up yield and property correlations) and Ri (parameter for conversion
PAloFCC, p e PAFCC, p e PAFCC, p " p ∈ PI FCC ð57Þ correlation), s∈SOFCC, j, and i are numerical identifiers.
4.2.2. Parameter Estimation and Model Validation. The
up
Vo,loFCC e Vo, FCC e Vo, FCC " o ∈ OFCC ð58Þ parameters in the empirical model for the FCC unit (βs,j and
ri) were estimated independently for each product yield and
property via the linear least-squares method (the correlations are
FF FCC , Fu0 , s00 , FCC , FCs , FFCC, s, u0 , FPFCC, s , X, X 0 , linear in the parameters) using MATLAB version 7.0.4. The
WHSV ∈ Rþ ; PAFCC, p , PCFCC, s, p , Vo, FCC ∈ R ð59Þ parity plots for product yield predictions (estimation and
validation) are presented in Figures 1012. In addition, the
Equations 31 and 32 calculate the total flow rate and the parity plot for the conversion is presented in Figure 13. The
relevant properties for the feed to the FCC unit. Equations coefficient of determination (R2) was found to be g0.950 for
3338 express the product yields (in units of wt %) for the FCC product yields and conversion correlations, and g0.964 for
products, as a function of feed properties and operating variables. product property correlations. The maximum and average
The conversion (in units of wt %), second-order conversion, and absolute errors for product yields and conversion correlations
weight hourly space velocity (WHSV) are estimated using eqs 39, are presented in Table 9. Table 9 shows that the absolute error
40, and 41, respectively. In eq 41, the scalar k is a conversion factor for product yield predictions was e2.09 wt % for parameter
(for WHSV) and the parameter CatInv represents the catalyst estimation and e1.14 wt % for model validation. Moreover, the
inventory in the reactor. Product properties are estimated using maximum and average absolute error for the coke yield predic-
eqs 4245 for naphtha, eqs 4650 for LCO, and eqs 5153 for tion using the same correlation presented in the UltraCat model
DO. The total flow rate for FCC products is estimated using eq 54. and fitting seven parameters (as in eq 38) were found to be
In addition, eq 55 denotes the volumetric balance at the product e0.63 wt % and e0.48 wt %, respectively. These values are
splitters. Inequalities 56 and 57 denote the capacity and feed greater than those obtained using the empirical correlation
property constraints for the FCC unit respectively. Inequality 58 proposed in this work (see Table 9). Similarly, the maximum
denotes the constraints for operating variables. Finally, the expres- and average absolute error for the conversion prediction using
sion described in eq 59 denotes the domain of the variables used in the same correlation presented in the UltraCat model and fitting
seven parameters (as in eq 39) were found to be e2.36 wt % and
e0.41 wt %, respectively. These values are close to those
obtained using the polynomial correlation proposed in this
work (see Table 9). Nevertheless, the polynomial correlations
presented in this work involve less-complex mathematical
expressions than the empirical correlations presented in the
UltraCat model.
Based on the results from parameter estimation and model
validation, it can be concluded that the empirical model for the
FCC unit is accurate and adequate for simulation of FCC
process. Finally, for the implementation of this empirical process
model (including empirical correlation for tail gases (H2S,
methane, ethane, etc.) not presented in this work), product
yield predictions should be normalized in order to ensure the
mass balance across the FCC unit, since mass balance was not
ensured for model prediction during the parameter estimation
phase. It is important to note that we proposed correlations
for LCO yield, DO yield, and conversion. Nevertheless, only
two of them should be used since the conversion is equal to
Figure 13. Parity plot for conversion prediction. 100  FCLCO  FCDO.

Table 9. Error for Parameter Estimation and Model Validation

Parameter Estimation Model Validation

product maximum absolute error (wt %) average absolute error (wt %) maximum absolute error (wt %) average absolute error (wt %)

C3T 0.25 0.09 0.21 0.07


C4T 0.51 0.15 0.46 0.16
NAP 2.09 0.41 1.14 0.45
LCO 0.58 0.15 0.40 0.15
DO 0.81 0.17 0.55 0.18
COK 0.22 0.08 0.21 0.09

conversion, X (wt %) 3.36 0.38 1.91 0.44

13415 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418


Industrial & Engineering Chemistry Research ARTICLE

5. CONCLUSIONS Up = units that can produce final products


In this work, empirical models for crude distillation units Uu = units u0 that can sent product streams as a feed to unit u
(CDUs) and a fluid catalytic cracking (FCC) unit were devel- USu0 ,u = product stream from unit u0 that can feed to unit u
oped and successfully validated using rigorous process simula-
Parameters
tors. In these models, product yields and properties are expressed Copu = operating cost for unit u
as nonlinear functions of operating variables and feed properties.
Cpru = price of final product from unit u
The operating variables are the reactor temperature and catalyst-
Crms = cost of raw material s
to-oil ratio for the FCC unit and the TBP cut points between
Demandu = demand for final product from unit u
adjacent fractions for CDUs. Several case studies have shown that
FFlou = lower bound for feed flow rate to unit u
the empirical models developed in this work can reproduce
FFup u = upper bound for feed flow rate to unit u
accurately the nonlinear refinery process behavior for given feed lo
PAu,p = lower bound of property p for the feed to unit u
properties and operating conditions. In addition, the empirical up
PAu,p = upper bound of property p for the feed to unit u
models presented in this work differ from those reported by
Supplys = availability of raw material s
Alhajri et al.22 mainly in two significant ways. First, the CDU
Vloo,u = lower bound for operating variable o of unit u
model was tuned using a rigorous process simulator that is Vup
o,u = upper bound for operating variable o of unit u
typically used in refinery operation for process monitoring.
Second, the FCC model proposed in this work considers more Variables
feed properties (density, sulfur content, and Conradson carbon FFu = feed flow rate to unit u
residue) and operating variables (reactor temperature and cata- FPu,s = flow rate of stream s from unit u
lyst-to-oil ratio) than those (characterization factor as feed Fu0 ,s,u = flow rate of stream s from unit u0 to unit u
property and conversion as operating variable) used by Alhajri PAu,p = property p of the total feed to unit u
et al.,22 which allows evaluation of the effect of some key variables PCu,s,p = property p of product stream s from unit u
on product yields and properties. Finally, the empirical models Vo,u = operating variable o of unit u
presented in this part are used in Part 2 of this work [Ind. Eng.
Chem. Res. 2011, DOI: 10.1021/ie200304v] for the production Empirical Model for CDUs
planning of petroleum refineries.
Indices
’ AUTHOR INFORMATION o = operating variable
p = property
Corresponding Author s = product stream from CDU
*E-mail: galoroux@usp.br. s0 = crude oil fed to the CDU
Tsi,si+1 = TBP cut point
Present Addresses
† r = hypothetical unit
Currently at Colombian Petroleum Research Institute (ICP-
u = unit
ECOPETROL), Colombia E-mails: omar@mail.pqi.ep.usp.br,
omarjo.guerra@ecopetrol.com.co. Sets
OCDU = operating variables for CDU
POCDU = relevant properties for product streams from CDU
’ ACKNOWLEDGMENT
POCDU,s = relevant properties for product stream s from CDU
The authors acknowledge the financial support from CNPq SOCDU = product streams from CDU
(Conselho Nacional de Desenvolvimento Científico e Tec- SUs,CDU = units u0 that can process product stream s from CDU
nologico, Brazil). USr,CDU = crude oils fed to CDU

’ NOMENCLATURE Parameters
Bias(s) = bias factor for the yield of product stream s from CDU
Planning Model (vol %)
FFloCDU = lower bound for feed flow rate to CDU
Indices FFupCDU = upper bound for feed flow rate to CDU
o = operating variable up 0
Fr,s 0
,CDU = availability of crude oil s fed to CDU
p = property lo
VTsi,si+1,CDU = lower bound for operating variable Tsi,si+1 of CDU
s = stream up
u = unit VTsi,si+1,CDU = upper bound for operating variable Tsi,si+1 of
u0 = unit CDU

Sets Variables
Ou = operating variables for unit u FCDU,s,u = flow rate of product stream s from CDU to unit u0
p = stream properties FDs,s = flow rate of product stream s from crude oil s0
PIu = relevant properties for the feed to unit u FFCDU = feed flow rate to CDU
POu,s = relevant properties for product stream s from unit u FPCDU,s = flow rate of stream s from CDU
SOu = product streams from unit u Fr,s0 ,CDU = feed flow rate of crude oil s0 to CDU
su = potential feed streams to unit u Mid_Vols,s = midvolume transfer ratio of product stream s from
SUs,u = units u0 that can process product stream s from unit u crude oil s0 (vol %)
Uf = units that can process external streams PCCDU,s,p = property p of product stream s from CDU

13416 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418


Industrial & Engineering Chemistry Research ARTICLE

PDs0 ,s,p = property p of product stream s from crude oil s0 PCFCC,LCO,VII = viscosity index of LCO from FCC unit
VDs0 ,Tsi,si+1 = volume percent vaporized of crude oil s0 at TBP cut PCFCC,NAP,API = API gravity of naphtha from FCC unit
point Tsi,si+1 (vol %) PCFCC,NAP,MON = motor octane number of naphtha from
VTsi,si+1,CDU = operating variable (TBP cut point) Tsi,si+1 for CDU FCC unit
PCFCC,NAP,RON = research octane number of naphtha from
Empirical Model for an FCC Unit FCC unit
PCFCC,NAP,SUL = amount of sulfur of naphtha from FCC unit
Indices (wt %)
i = numeric identifier for parameters in product yields and PCFCC,s,SPG = specific gravity of product stream s from FCC unit
properties correlations PCu0 ,s00 ,p = property p of stream s00 from unit u0 to FCC unit
j = numeric identifier for parameters in conversion correlation VCO,FCC = catalyst-to-oil ratio in the FCC unit
o = operating variable VTrxn,FCC = reactor temperature in the FCC unit (F)
p = property WHSV = weight hourly space velocity in the FCC unit (h1)
s = product stream from FCC unit X = weight conversion in the FCC unit (wt %)
s00 = stream feed to the FCC unit X0 = second-order conversion in the FCC unit
u0 = unit

Sets
OFCC = operating variables for FCC unit ’ REFERENCES
PIFCC = relevant properties for the feed to FCC unit (1) Parkash, S. Refining Processes Handbook; Elsevier: Amsterdam,
POFCC,LCO = relevant properties for light cycle oil from FCC unit 2003.
POFCC,NAP = relevant properties for naphtha from FCC unit (2) Moro, L. F.; Zanin, A. C.; Pinto, J. M. A planning model for
refinery diesel production. Comput. Chem. Eng. 1998, 22 (Suppl. 1),
POFCC,DO = relevant properties for decanted oil from FCC unit
S1039–S1042.
SOFCC = product streams from FCC unit (3) Watkins, R. Petroleum Refinery Distillation, 2nd ed.; Gulf Publishing
SUs,FCC = units u0 that can process product stream s from Company: Houston, TX, 1979.
FCC unit (4) Hartmann, J. C. Distinguish between scheduling and planning
UFCC = units u0 that can sent product streams as a feed to models. Hydrocarbon Process. 1998, 77 (7), 93–100.
FCC unit (5) Kreipl, S.; Pinedo, M. Planning and Scheduling in Supply
USu0 ,FCC = product stream from unit u0 that can feed to FCC unit Chains: An Overview of Issues in Practice. Prod. Oper. Manage. 2004,
13 (1), 77–92.
Parameters (6) Erdirik-Dogan, M.; Grossmann, I. E. A Decomposition Method
Ri = parameter i for conversion (X) correlation for the Simultaneous Planning and Scheduling of Single-Stage Contin-
βs,j = parameter j for product yields and properties correlations uous Multiproduct Plants. Ind. Eng. Chem. Res. 2006, 45 (1), 299–315.
CatInv = catalyst inventory in the FCC reactor (lb) (7) Erdirik-Dogan, M.; Grossmann, I. E. Simultaneous planning and
k = conversion factor scheduling of single-stage multi-product continuous plants with parallel
lines. Comput. Chem. Eng. 2008, 32 (11), 2664–2683.
FFloFCC = lower bound for feed flow rate to FCC unit
(8) Maravelias, C. T.; Sung, C. Integration of Production Planning
FFup FCC = upper bound for feed flow rate to FCC unit and Scheduling: Overview, Challenges and Opportunities. In Proceed-
PAloFCC,p = lower bound of property p for the feed to FCC unit ings, Foundations of Computer-Aided Process Operations (FOCAPO
PAup FCC,p = upper bound of property p for the feed to FCC unit 2008); CACHE Corporation: Cambridge, MA, 2008; pp 1322.
Vloo,FCC = lower bound for operating variable o of FCC unit (9) Pinto, J. M.; Joly, M.; Moro, L. F. Planning and scheduling models
Vup
o,FCC = upper bound for operating variable o of FCC unit for refinery operations. Comput. Chem. Eng. 2000, 24 (910), 2259–2276.
(10) Kelly, J. D. Formulating Production Planning Models. Chem.
Variables Eng. Prog. 2004, 100 (1), 43–50.
FCCOK = coke yield in the FCC unit (wt %) (11) Li, J.; Wenkai, L.; Karimi, I. A.; Srinivasan, R. Improving the
FCC3T = C3s yield in the FCC unit (wt %) Robustness and Efficiency of Crude Scheduling Algorithms. AIChE J.
FCC4T = C4s yield in the FCC unit (wt %) 2007, 53, 2659–2680.
FCDO = decanted oil yield in the FCC unit (wt %) (12) Riazi, M. R. Characterization and Properties of Petroleum Frac-
tions; ASTM International: West Conshohocken, PA, 2005.
FCLCO = LCO yield in the FCC unit (wt %)
(13) Zhang, J.; Zhu, X.; Towler, G. A Level-by-Level Debottleneck-
FCNAP = naphtha yield in the FCC unit (wt %) ing Approach in Refinery Operation. Ind. Eng. Chem. Res. 2001,
FFCC,s,u0 = flow rate of product stream s from FCC unit to unit u0 40, 1528–1540.
FFFCC = feed flow rate to FCC unit (14) Trierwiler, L. Advances in crude oil LP modeling. Presented at
FPFCC,s = flow rate of product stream s from FCC unit the NPRA Computer Conference, October 13, 2001, Dallas, TX.
Fu0 ,s00 ,FCC = feed flow rate of stream s00 from unit u0 to FCC unit (15) Li, W.; Hui, C.-W.; Li, A. Integrating CDU, FCC and product
PAFCC,p = property p of the total feed to FCC unit blending models into refinery planning. Comput. Chem. Eng. 2005, 29
PAFCC,SPG = specific gravity of total feed to FCC unit (9), 2010–2028.
PCFCC,DO,API = API gravity of decanted oil from FCC unit (16) Robinson, P. R. Petroleum Processing Overview. In Practical
PCFCC,DO,SUL = amount of sulfur of decanted oil from FCC unit Advances in Petroleum Processing, Vol. 1; Hsu, C. S., Robinson, P. R.,
Eds.; Springer Science: New York, 2006; pp 178.
(wt %)
(17) Sadeghbeigi, R. Fluid Catalytic Cracking Handbook, 2nd ed.;
PCFCC,DO,VII = viscosity index of decanted oil from FCC unit Gulf Publishing Company: Houston, TX, 2000.
PCFCC,LCO,API = API gravity of LCO from FCC unit (18) Wollaston, E.; Haflin, W.; Ford, W.; D’Souza, G. What influ-
PCFCC,LCO,CTI = cetane index of LCO from FCC unit ences cat cracking. Hydrocarbon Process. 1975, 54 (9), 93–100.
PCFCC,LCO,PPI = pour point index of LCO from FCC unit (19) Gary, J. H.; Handwerk, G. E. Petroleum Refining Technology and
PCFCC,LCO,SUL = amount of sulfur of LCO from FCC unit (wt %) Economics, 4th ed.; Marcel Dekker: New York, 2001.

13417 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418


Industrial & Engineering Chemistry Research ARTICLE

(20) Jacob, S. M.; Gross, B.; Weekmann, V. W. A Lumping and


Reaction Scheme for Catalytic Cracking. AIChE J. 1976, 22 (4),
701–713.
(21) Hernandez-Barajas, J. R.; Vazquez-Roman, R.; Felix-Flores, M.
A comprehensive estimation of kinetic parameters in lumped catalytic
cracking reaction models. Fuel 2009, 88 (1), 169–178.
(22) Alhajri, I.; Elkamel, A.; Albahri, T.; Douglas, P. L. A Nonlinear
Programming Model for Refinery Planning and Optimisation with
Rigorous Process Models and Product Quality Specifications. Int. J.
Oil Gas Coal Technol. 2008, 1 (3), 283–307.

13418 dx.doi.org/10.1021/ie200303m |Ind. Eng. Chem. Res. 2011, 50, 13403–13418

Potrebbero piacerti anche