European Symposium on Computer Aided Process Engineering - 13

A. Kraslawski and I. Turunen (Editors)

© 2003 Elsevier Science B.V. All rights reserved. 623

Multiobjective Optimisation of Fluid Catalytic Cracker

Unit Using Genetic Algorithms
Dhaval Dave and Nan Zhang
Department of Process Integration,
UMIST, Manchester, UK, M60 IQD

Abstract
A rigorous model is presented for simulating an industrial fluid catalytic cracker unit.
Two different kinetic lumping schemes are presented for FCC reactor modelling and the
regenerator is modelled as two separate regions: the dense bed and the dilute phase.
Kinetic parameters of five lump model were tuned with industrial data. An adapted
version of nondominated sorting genetic algorithm (NSGA) was then used to optimise
the performance of the unit. Operational insights are developed by using several
objective functions and decision variables are obtained for optimal operation.
Capabilities of an algorithm are presented by case study using five lump kinetic scheme
and objectives considered are maximisation of gasoline production and minimisation of
CO emission from regenerator. More detailed ten lump kinetic scheme is proposed for
multi objective analysis of recycle slurry flow and CO emission from regenerator.
Pareto optimal solutions are obtained and the results are expected to enable the process
engineer to gain useful insights to locate compromised operating conditions.

1. Introduction
Fluid Catalytic Cracking (FCC) is one of the most important processes in an oil
refinery. Its function is to convert heavy hydrocarbon petroleum streams into more
valuable, lighter hydrocarbon fractions such as middle distillate, gasoline and liquefied
petroleum gas etc. Optimal operation of the process is decisive for overall economic and
environmental health. Often several objectives and constraints are involved in the
process, and optimisation studies incorporating these conflicting objectives would be
invaluable to the process engineer.
Numerous papers relating to the FCC process can be found in the published technical
literature. They present various aspects of design, kinetics, mathematical modelling and
simulation, stability, optimisation and control. The history of the development and
commercialisation of catalytic cracking was reconstructed in detail in a review by
Avidan and Shinnar (1990). Different workers have discussed the kinetics in the reactor
and the regenerator and have modelled these units separately, while others have
developed integrated models for the reactor-regenerator system. Several studies have
also been carried out on optimisation and optimising control of FCC units. Most of the
studies in optimisation of FCCU were limited to single objectives (e.g. profit
maximisation, maximisation of conversion, maximisation of individual product
production etc.).
In the present work, multiobjective optimisation study is performed on an industrial
fluid catalytic cracker unit. We use an adapted version of nondominated sorting genetic
624

algorithm (NSGA) (Deb and Srinivas, 1995) and the objective functions considered in
the present study are maximisation of gasoline production and minimisation of CO
emission from regenerator. The objective functions considered for more detailed kinetic
model are to maximise recycle slurry flow and to minimise CO emission from
regenerator (and/or maximisation of gasoline production).

2. FCCU Modelling
The FCC unit comprises of two basic parts, a reactor/riser in which hydrocarbon
cracking reactions occur and a regenerator in which catalyst regains its activity by
burning coke deposited on it during cracking. More detailed description of the process
is available in Avidan and Shinnar (1990). Recently, Dave and Saraf (2002) reviewed
the extensive literature available on modelling of industrial FCC units.
Selection of an appropriate lumping scheme was one of the most important issues in this
modelling exercise. Ten lump kinetic scheme developed by Jacob et al. (1976) and five
lump kinetic model proposed by Ancheyta et al. (1999) were examined closely. The
virtue of more detailed lumping scheme over other less detailed models is that rate
constants is that rate constants are independent of feed composition. But utilisation of
these models are limited by two problems i.e. detailed characterisation of streams is not
available on a regular basis and elaborate kinetic information is scarcely available.
Thus, a balance between kinetic description required and cost of laboratory analysis
often decides selection of lumping strategy.

2.1. Five-lump model

Dave and Saraf (2002) modified the original scheme of Ancheyta et al. (1999) by
assuming gasoline and LPG also convert to coke. Figure 1 presents the modified kinetic
study, used in the present work. Since rate constants for this model are dependent upon
feed quality, they are obtained by tuning industrial data available. In the present work,
five lump kinetic model developed by Dave and Saraf (2002) is used for multi objective
analysis for objective functions gasoline production maximisation versus minimisation
of CO emission from regenerator.

2.2. Ten-Lump model

Since five-lump model is unable to capture feed quality and hence impact of recycle
slurry flow rate on the performance of unit. It was necessary to develop a model, which
is based upon detailed characterisation of feed. In present study.
Ten lump kinetic scheme (as shown in figure 2) developed by Jacob et al. (1976) is
used with some modifications. Since Jacob et al. (1976) lumped coke and gas together,
regenerator model can not be integrated with its present form. In this case, gas yield was
predicted from correlation available in Gary and Hadwerk (1994) and hence ten lump
model was made suitable. This model has also been tuned for an industrial application.
Since hydrodynamics and detailed design were idealised, it was necessary to tune the
model to match with industrial performance.
Dave and Saraf (2002) described regenerator as two-region (the dense bed and the dilute
phase) model, essentially following the scheme of Krishna et al. (1985) with some
modification. Bulk of the coke combustion reaction occurs in the dense bed and the
dilute phase is the region above the dense bed, 'after burning' of carbon monoxide and
catalyst entrainment are the main effects of this section. The dense bed is modelled as a
 625

well-mixed reactor with regard to catalysts and a plug flow reactor for gas stream,
whereas dilute phase is modelled as a plug flow reactor for both entrained catalyst
particles and gas stream.

ki

^ m*
GaiOil^

• Coke 4 "

Figure 1 Five lump kinetic scheme Figure 2 Ten lump kinetic scheme
(Dave and Saraf, 2002). (Jacob etal. 1976).

The model involves many ordinary differential equations which are non-stiff and solved
using Runge-Kutta-Gill method (Gupta, 1995). Integrated reactor and regenerator model
was solved using Newton-Raphson method (Gupta, 1995). The complete set of the
reactor and the regenerator model equations as well as the values of the associated
parameters are given in Dave and Saraf (2(X)2).

3. Multiobjective Optimisation
The operation of the FCC unit above is now optimised. It was assumed that economics
of FCC unit is affected by Ffeed, Tfeed, Feat, Tair and Fair- Furthermore, it is assumed that
the regenerator works in full combustion mode and there is no CO- reboiler present.
Many refiners prefer to work on full combustion mode because of regeneration effect.
Many objective functions can be considered in an optimisation study. In the given unit
it was imperative to maximise gasoline production, so it was chosen as one of the
objective. Maximisation of gasoline requires more conversion and more throughput, and
hence more amount of coke formation. This coke deactivates the catalyst and therefore
it is require burning off as much coke as possible in the regenerator, requiring higher
amount of air. Burning of coke results in the formation of pollutants CO and CO2. The
regenerator size is fixed and also there is a limited capacity of regenerator feed air
blower. So as to increase the gasoline production beyond certain extent, trade-off is
generally made by operating regenerator in the partial combustion mode (i.e. emitting
more CO through regenerator). Also refiners would like to increase recycle slurry flow
rate because unconverted heavy slurry is the lowest price stream. Increasing the recycle
flow has an adverse impact on coke formation and hence to maintain conversion levels
at desired level, trade off is made by partial conversion. Based on above discussion, we
could see that objectives involved here are gasoline production maximisation
(maximisation of gasoline yield + throughput), minimisation of regenerator air flow
rate, minimisation of CO emission and maximisation of recycle slurry flow rate.
626

Consideration of all the objectives simultaneously is difficult to analyse. So we form

three problems which optimise two objectives simultaneously for simplicity.
The optimisation problems are solved using genetic algorithm (GA) made suitable for
multiobjective problems and known as NSGA. NSGA differs from traditional GA in the
sorting procedure adapted. The potential solutions are sorted twice, first nondominated
solutions are obtained and these are then sorted depending upon the number of other
chromosomes nearby. Thus, the algorithm is able to fmd solutions closer to the true sets
of compromised solutions and maintains diversity in the gene pool at each time. Details
of this algorithm are provided by Deb and Srinivas (1995).

3.2. Problem 1 (Maximisation of gasoline yield vs. minimisation of CO emission)

M a x F i ( X ) = Yconversion (1)

^conversion = sum of the yiclds of gasoline, LPG, dry gas and coke.
It should be noted that using five lump model (Dave and Saraf, 2002), maximisation of
conversion is equivalent to maximisation of gasoline conversion.

Max F2(X) = 10.0/ (1 + Xco) (2)

Where decision variables X = {Tfeed. Feat»Fair»Tair }*

Bounds on the decision variables are specified such that it captures operations of all
industrial FCC units. It should be emphasised here that the total feed flowrate Ftot (Ffresh
+ Frecy) is fixcd at value 28 kg/s. The constraints other than the model equations are as
follows.

Trgn < 1000 K (3)

Crgn < 0.001 (kg coke/kg catalyst) (4)

Constraints on regenerator temperature and coke content of regenerated catalyst were

required in order to prevent catalyst deactivation due to thermal damage and excessive
coke deposition. The Pareto optimal solutions obtained along with the decision
variables corresponding to each point on the pareto curve is given in figure 3.

3.3. Problem 2 (Maximisation of gasoline production vs. Minimisation of CO

emission)

MaxFi(X)=Ftot*(Ygasoline) (5)

Max F2(X) = 10.0/ (1 + Xco) (6)

W h e r e X = { Ftot, Tfeed, Feat y Fair , Tair }

* Tfeed = Feed preheat temperature, Feat = Catalyst flowrate , Fair = air flowrate to the regenerator, Tair =
Regenerator air preheat temperature, Ffresh = Fresh feed flowrate, Frecy = Recycle feed flowrate, Ftot = Total feed
flow rate, Xco = Composition of Carbon monoxide in flue gas (vol%)
 627

It should be noted that again five lump model (Dave and Saraf, 2002) is used. It was
assumed that composition of feed is maintained constant by fixing recycle ratio.
Appropriate bounds on the decision variables are specified and the constraints for this
problem are same as that are listed in problem 1.

3.4. Problem 3 (Maximisation of recycle slurry flowrate vs. maximisation of

gasoline production)

M a x F i ( X ) = (Ffresh+ Frecy)*(Ygasoline) (7)

M a x F 2 ( X ) = Frecy (8)

W l i e r e A = | Frecy? Afresh? ^ fegj. Teat ? ^air ? ^ aj^ )

Modified ten lump model is used, since feed composition is allowed to change in this
problem. Appropriate bounds on the decision variables are specified and the constraints
specified in problem 1 are applied. Also an additional constraint is specified to ensure
CO emission within some limit.

Xco<2% (9)

It is because of limitation of space pareto optimal solutions obtained for problem 2 and
3 are not shown. The results are expected to enable the process engineer to gain useful
insights and to locate compromised operating conditions. In fact, such procedure can be
applied for many other objectives for FCCU.
.„». 4.6 24
^o
1
QOM 4.5
23
>
c
QGCG5

Q003
*
4.4
4.3 / i ^Vt^^ 22
?1
•
•
"O 20
o CXCE5
*
4? • \1 19
w
40
QOGB

QCOB
^^ \1 «
4.1
•
• i
18
17
» • u. •
E QOOt
o
39
> i 16
LU CXXXE •;

38 15
•
0 o b «»
80 65 70
" is 6D & ^ ^' * ?tt 7!2
o eD ee 64 68 68 7D 72

o Conversion Conversion Conversion

6B0
530
640 ^f^lMl
525
eao
e 520
515
*^% *^ * *40^ 620 a^ *
-a 510
610
505'
eoo • • • | l *
500
495 590

490 580
D 62 64 66 68 70 72 eb 62 'M '^ 68 7D 72

Converelon Conversion Conversion

Figure 3 Pareto optimal solutions obtained for problem 1 using five lump model.
628

4. Conclusions
Two different kinetic lumping models are tuned in order to simulate an industrial FCC
unit. Operational insights are developed by performing multiobjective optimisation
study using non dominated sorting genetic algorithm. Pareto optimal solutions are
obtained for different objective functions and constraints considered, which are
expected to help process engineer to locate favoured solution.

5. References
Ancheyta, J.J.; Lopez, I.F; Aguilar, R.E.; Moreno, M.J., 1997, A Strategy for Kinetic
Parameter Estimation in the Fluid Catalytic Cracking Process, Ind. Eng. Chem.
Res., 36, 5170-5174.
Avidan, A.A.; Shinnar, R., 1990 Development of Catalytic Cracking Technology. A
Lesson in Chemical Reactor Design, Ind. Eng. Chem. Res., 29, 931-942.
Dave, D.J. and Saraf, D.N., 2002, A model suitable for rating and optimization of
industrial FCC units, selected for publication in Indian Chemical Engineer.
Deb, K. and Srinivas, N., 1995, Multiobjective optimizatoin using nondominated
sorting in genetic algorithm Evol. Comput., 2, 106-114.
Gary, J.H. and Handwerk, G.E., 1993, Petroleum refining, technology and economics,
3, Marcel Dekker.
Gupta, S.K., 1995, Numerical Methods for Engineers, Wiley Eastern/New Age Intl.
Jacob, S.M.; Gross, B.; Voltz, S.E.; Weekman, V.M., Jr., 1976, A Lumping and
Reaction Scheme for Catalytic Cracking, AIChE J., 22(4), 701-713.
Krishna, A.S. and Parkin, E.S., 1985, Modeling the Regenerator in Commercial Fluid
Catalytic Cracking Units, Chem. Eng. Prog., 81(4), 57-62.