9th International Symposium on Advanced Control of Chemical Processes

9th
JuneInternational Symposium
7-10, 2015. Whistler, on Advanced
British Columbia,Control
Canadaof Chemical Processes
9th International
9th International Symposium
Symposium on Advanced
on Advanced Control
Control of Chemical
of Chemical Processes
Processes
June 7-10, 2015. Whistler, British Columbia, Canada
Available online at www.sciencedirect.com
June 7-10, 2015. Whistler, British Columbia, Canada
June 7-10, 2015. Whistler, British Columbia, Canada

ScienceDirect
IFAC-PapersOnLine 48-8 (2015) 373–378
Optimization
Optimization of
of Catalytic
Catalytic Naphtha
Naphtha
Optimization
Reforming of
ProcessCatalytic
Based onNaphtha
Modified
Reforming
Reforming Process
Process Based
Based on
on Modified
Modified
⋆
Differential Evolution Algorithm
Differential Evolution Algorithm ⋆⋆
Differential Evolution Algorithm
Min Wei ∗∗ Minglei Yang ∗∗ Feng Qian ∗∗ Wenli Du ∗∗
Min
Min Wei ∗∗ Minglei Yang ∗∗ Feng Qian ∗∗ Wenli Du ∗∗
Min Wei Wei MingleiMinglei Yang Yang Feng Feng QianQian Wenli Wenli Du Du
∗
∗ Key Laboratory of Advanced Control and Optimization for Chemical
∗ Key
∗ Key
Laboratory
Process, Ministry
Laboratory of Advanced
of of Education,
Advanced Control and
Shanghai,
Control Optimization
CO 200237, for
and Optimization
Optimization for
PRC
for Chemical
(Tel:
Chemical
Key Laboratory
Process, Ministry of Advanced
of Education, Control and
Shanghai, CO 200237, PRC Chemical
(Tel:
Process, 13701941965;
Process, Ministry
Ministry of
of e-mail:gydm890622@126.com)
Education,
Education, Shanghai,
Shanghai, CO
CO 200237,
200237, PRC
PRC (Tel:
(Tel:
13701941965;
13701941965; e-mail:gydm890622@126.com)
e-mail:gydm890622@126.com)
13701941965; e-mail:gydm890622@126.com)
Abstract: Catalytic naphtha reforming is one of the most important processes for high
Abstract:
octane
Abstract: gasoline Catalytic naphtha
manufacture
Catalytic naphtha andreforming is one of
aromatic hydrocarbons
reforming is the production.
most important
In thisprocesses for
article, a modified high
Abstract:
octane gasoline Catalytic naphtha
manufacture and reforming
aromatic is one
one ofof the
hydrocarbons
most
most important
the production.important
In this
processes
processes
article, a
for
for high
high
modified
differential
octane
octane gasoline evolution
gasoline (DE) algorithm
manufacture
manufacture and
and aromatic
aromatic is proposed to optimize
hydrocarbons
hydrocarbons an actual
production.
production. In
In thiscontinuous
this article, aa catalytic
article, modified
modified
differential
naphtha
differential evolution
reforming
evolution (DE)
(CCR) algorithm
(DE)process.
algorithm Theis is proposed
isoptimization
proposed to to optimize
toproblem an
optimizeconsiders actual
an actual
actual continuous
to continuous
minimize the
continuous catalytic
energy
catalytic
differential
naphtha evolution
reforming (DE)
(CCR) algorithm
process. The proposed
optimization optimize
problem an
considers to minimize the catalytic
energy
consumption
naphtha
naphtha and maximize
reforming
reforming (CCR)
(CCR) the aromatics
process.
process. The
The yield. The
optimization
optimization CCR process
problem
problem considers
considers model
to
to is established
minimize
minimize the
the by
energy
energy
consumption
adopting
consumption the and
and maximize
27-lumped
maximizekineticsthe
the aromatics
reaction
aromatics yield.
network,
yield. The
and
The CCR
all
CCR process
parameters
process model
are is
adjusted
model is established
based
established on by
the
by
consumption
adopting the and maximize
27-lumped kineticsthe reaction
aromatics yield. and
network, Theall CCR process are
parameters model is established
adjusted based on by
the
actual
adopting
adopting process
the data.
27-lumped
the 27-lumped The DE
kinetics
kineticsalgorithm
reaction is modified
network,
reactionisnetwork, to
and
and maintain
all the
parameters
all parameters diversity
are of
adjusted
are adjusted the population.
based
based on the
on the
actual
In
actual process
thisprocess
mechanism,
process data.
data. TheThe
The DE
individuals algorithm
DE algorithmfurtheris
algorithm modified
isfrom to
the best
modified maintain
individual
to maintain
maintain the diversity
the have of
larger
diversity the population.
possibilities
of the
the population. to
actual
In this mechanism,data. DE
individuals further modified
from the to
best individual the diversity
have of
larger population.
possibilities to
be
In
In selected
this
this in
mechanism,
mechanism,the mutation
individuals
individuals operator.
further
furtherTheThe from
frommodified
the DE
best
the best is evaluated
individual
individual have
haveby solving
larger
larger 6 benchmark
possibilities
possibilities to
to
be
be selected
functions,
selected andin
in the
the mutation
performance
mutation operator.
is compared
operator. The modified
with
modified DE
classic
DE is
DEs.
is evaluated
The
evaluated by
results
by solving
demonstrate
solving 6
6 benchmark
that
benchmark the
be selected
functions, andin the
the mutation
performance operator.
is The modified DE is evaluated by solving 6 benchmark
modified
functions,DE
functions, andhas
and thebetter
the performance
performance is compared
global search
is ability and
compared
compared
with
with
classic
withhigher
classic
DEs.
DEs. The
DEs. The results
classiccomputation
The results
results
demonstrate
efficiency. Furthermore,
demonstrate
demonstrate
that the
that the
that the
modified
optimization
modified DE DE
DE has has better
results
has better of global
catalytic
better global search
naphtha
global search ability
search ability
ability andand
reforming higher computation
process
and higher indicate
higher computation efficiency.
that
computation efficiency. the Furthermore,
proposed
efficiency. Furthermore, algorithm
Furthermore, the the
the
modified
optimization
has the abilityresults
optimization
optimization results
results
of
of locating
of catalytic
of catalytic
catalytic
naphtha
the optimal
naphtha
naphtha
reforming
operating
reforming
reforming
process
points,
process
process
indicate
in which the that
indicate
indicate that
that
the
aromatics
the proposed
the proposed
proposed
algorithm
yield is improved,
algorithm
algorithm
has
while
has the
the ability
energy
ability of
of locating
consumption
locating the
theis optimal
reduced.
optimal operating
Meanwhile,
operating points,
the
points, in which
optimal
in which the aromatics
operating
the points
aromatics yield
and
yield is
is improved,
results
improved,are
has theenergy
while abilityconsumption
of locating the is optimal operating
reduced. Meanwhile, points,
the in whichoperating
optimal the aromaticspointsyield
andisresults
improved,
discussed
while energy
while energy at the end of theis
consumption
consumption isarticle.
reduced. Meanwhile,
reduced. Meanwhile, the the optimal
optimal operating
operating points
points and results are
and results are
are
discussed
discussed at
at the
the end
end of
of the
the article.
article.
discussed at the end of the article.
© 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Keywords: Catalytic naphtha reforming process, Energy consumption, Differential evolution,
Keywords:
Global
Keywords: searchCatalytic naphtha
Catalytic naphtha reforming process, process, Energy consumption,
consumption, Differential evolution, evolution,
Keywords:
Global searchCatalytic naphtha reforming reforming process, Energy Energy consumption, DifferentialDifferential evolution,
Global
Global search
search
1. INTRODUCTION R ecycle G as
1.
1. INTRODUCTION R ecycle G as

1. INTRODUCTION
INTRODUCTION R ecycle G as
R ecycle G as

Continuous catalytic reforming of naphtha (CCR) process

Continuous
plays catalyticrole
a significant
Continuous catalytic reforming
for high
reforming of naphtha
of naphtha (CCR) process
octane gasoline
(CCR) process
manu-
Continuous
plays a catalyticrole
significant reforming
for highof naphtha
octane (CCR) process
gasoline manu-


facture
plays
plays aa and aromaticrole
significant
significant hydrocarbons
role for
for high
high production.
octane
octane gasoline
gasoline The sim-
manu-
manu-


facture and aromatic hydrocarbons production. The sim-











plified
facture process
and flow diagram
aromatic hydrocarbonsof thisproduction.
process is shown The as
sim-










facture
plified and
process aromatic
flow hydrocarbons
diagram of this production.
process is The
shown sim-
as
















Fig.1.
plified
plified In the
process
process CCRflow process,
diagram
flowprocess,
diagram aromatic
of this yield
process
of this yield
processis anis important
shown
is important
shown as as
Fig.1. In
assessment the CCR aromatic is an
Fig.1.
Fig.1. In
assessment thestandard
In the CCR
CCR process,
standard
of thearomatic
process,
of the
the
production
aromatic
production
yield
yield quality.
is
is an However,
an important
quality. important
However,
improving
assessment
assessment the aromatics
standard
standard of
of yield
the is often achieved
production
production quality.
quality. atHowever,
the cost
However, R eform ate
improving
of larger
improving the
material
the aromatics
and
aromatics energyyield
yield is often
consumption.
is often achieved at
Sometimes,
achieved at the cost
the
at the cost
the
R eform ate

improving
of larger the aromatics
material and energyyieldconsumption.
is often achieved Sometimes, cost
the    
R eform ate
R eform ate

losses
of
of caused
larger
larger by
material
material improving
and energy
and energy the aromatics
consumption.
consumption. yield is even
Sometimes,
Sometimes, larg-
the
the    

losses
er thancaused by
the benefits improving the
it produced. aromatics yield
Therefore, is even larg- 








losses caused by
by improving the
the aromatics yieldoptimization
yield is
is even
even larg-  
losses
er thancaused
the improving
benefits it produced. aromatics
Therefore, optimization larg- 
N aphtha Feed 


on
er
er the CCR
than the process it
the benefits is of important
produced. significance.
Therefore, optimization 


on than
the CCR
CCR benefits
process it produced.
is of
of important Therefore,
significance. optimization 
N aphtha Feed 
N aphtha Feed 
on
on the
the CCR process
process is
is of important
important significance.
significance.
N aphtha Feed

Differential evolution (DE) is an efficient optimization al- Fig. 1. Simplified process flow diagram of the CCR process
Differential
gorithm,
Differential whichevolution (DE) is
offers advantages
evolution (DE) is an over
efficient optimization
traditional population-al- Fig. 1. Simplified process flow diagram of the CCR process
Differential
gorithm, whichevolution
offers (DE) is an
advantages
efficient
an over
efficient optimization
optimization
traditional population-
al-
al- Fig. 1. Simplified process flow diagram of the CCR process
based
gorithm,
gorithm, stochastic
which
which algorithm
offers
offers advantages
advantages in convergence
over
over traditional
traditional and ro- Fig. 1. Simplified process flow diagram of the CCR process
speedpopulation-
population-
based stochastic
stochastic algorithm in (2011).
convergence speed amount
and ro- ro-
bustness,
based Das andalgorithm
based stochastic Suganthan
algorithm in
in (2011). In decades,
convergence
convergence speed and ro- et
and
speed amount et
al. (2010) used the DE method to solve dynamic opti-
al. (2010) used the DE method to solve dynamic
bustness,
of research
bustness,
bustness,
Das
Das
Das
and
works
and
and
Suganthan
on the applications
Suganthan
Suganthan (2011).
(2011).
In
In
In
decades,
of DE algorithm
decades,
decades, amount
amount mization
et
et al. problem
al. (2010)
(2010) used of
used thethe
the DEnaphtha
DE methodreforming
method to
to solve process. opti-
solve dynamic
dynamic opti-
opti-
of research works on the applications of DE algorithm mization
mization problem
problem of
of the
the naphtha
naphtha reforming
reforming process.
process.
in
of
of chemical
research
research processes
works
works on
on have
the
the been
applications
applicationsproposed.
of
of DE
DE Angira
algorithm
algorithm and mization problem of the naphtha reforming process.
in chemical
chemical
Santosh (2007) processes have
applied have been proposed.
trigonometric proposed.
mutation Angira
operation and Despite that many research works have been investigated
in
in chemical
Santosh (2007)
processes
processes
applied have been
been proposed.
trigonometric mutation
Angira
Angira
operation and Despite
and on
Despite
Despite
that
that many research
the applications
that many of modified
many research
research
works
DEs,
works
works
have
most
have
have
been investigated
of them
been
been mainly
investigated
investigated
in the
Santosh
Santosh DE algorithm,
(2007)
(2007) applied
applied andand the
trigonometric
trigonometric updated algorithm
mutation
mutation were
operation
operation on the
focused
on the applications
on the
applications of
discovery
of modified
of
modified DEs,
novel
DEs, most
mutation
most of
of them mainly
operators
them mainlyor
in
in the
utilized
the DE
DE algorithm,
to solve seven cases
algorithm, and the
of
the updated
chemical
updated algorithm
process
algorithm were
optimiza-
were on the applications
focused on the of modified
discovery of DEs,
novel most of them
mutation mainly
operators or
in the
utilized DE
to algorithm,
solve seven and
cases the
of updated
chemical algorithm
process were
optimiza- the novel
focused
focused on settings
the
onsettings of
the discovery operating
discovery of parameters.
novel
of novel mutation
mutation The The problem
operators
operators or
or
tion problems.
utilized
utilized to
to solve
solve Xu et al.
seven
seven (2013)
cases
cases of
of proposedprocess
chemical
chemical an improved
process optimiza-
optimiza- the the of novel
population
the novel
novel settings of
diversity
settings of operating
decreasing
of operating parameters.
operating parameters.has not
parameters. The beenproblem
solved
The problem
problem
tion
DE
tion problems.
algorithm
problems. Xu
based
Xu et
et al.
on
al. (2013)
(2013) proposed
self-adaptive
proposed an
strategy.
an improved
Rahimpour
improved the
the the
of population diversity decreasing has not been solved
tion problems.based
DE algorithm
algorithm Xu eton al.self-adaptive
(2013) proposed an improved
strategy. Rahimpour the effectively.
of
of population
population Thisdiversity
defect may
diversity result inhas
decreasing
decreasing hasrepetitive
not
not been counting
been solved
solved
DE
DE algorithm based based on on self-adaptive
self-adaptive strategy. Rahimpour effectively.
strategy. Rahimpour and weak
effectively.
effectively.
This
global
This
This
defect
search
defect
defect
may
may
may
result
ability, which
result
result
in
in
in
repetitive
are counting
unacceptable
repetitive
repetitive counting
countingin
⋆ This work is supported by Major State Basic Research Develop- and
and weak
solving
weak global
optimization
global search
search ability,
problems
ability, which
of the
which are unacceptable
chemical
are in
processes.
unacceptable in
⋆ This work is supported by Major State Basic Research Develop- and weak
solving global search
optimization ability, of
problems which
the are unacceptable
chemical in
processes.
⋆
ment
⋆ Program
This work is of China (2012CB720500),
supported by Major State National
Basic NaturalDevelop-
Research Science In this
solving work,
solving optimizationa modification
optimization problems strategy
problems of of the on DE
the chemical algorithm
chemical processes.
processes.is
ThisProgram
ment work is of
Foundation of
supported
China
China
by Major State National
(2012CB720500),
(U1162202), and 111
Basic Research
Project Natural
(B08021).
Develop-
Science In
In this
proposed.
this work,
This
work, a
a modification
method adopts
modification strategy
distance
strategy on
on DE algorithm
information
DE of
algorithm is
the
is
ment Program
ment Program of of China
China (2012CB720500),
(2012CB720500), National Natural Natural Science In this work, a method
modification strategy oninformation
DE algorithm is
Foundation of China (U1162202), and 111 National
Project (B08021).Science proposed. This adopts distance of the
Foundation of China (U1162202), and 111 Project
Foundation of China (U1162202), and 111 Project (B08021). (B08021). proposed. This method adopts distance
proposed. This method adopts distance information of theinformation of the
Copyright
2405-8963 ©© 2015,
2015 IFAC
IFAC (International Federation of Automatic Control)373 Hosting by Elsevier Ltd. All rights reserved.
Copyright © 2015 IFAC 373
Peer review
Copyright © under
2015 responsibility
IFAC
Copyright © 2015 IFAC of International Federation of Automatic
373Control.
373
10.1016/j.ifacol.2015.08.210
IFAC ADCHEM 2015
374
June 7-10, 2015. Whistler, BC, Canada Min Wei et al. / IFAC-PapersOnLine 48-8 (2015) 373–378

population to determine the possibilities that individuals   

would be chosen in the mutation operator.
    
This work is organized as follow: Section 2 describes the
CCR process in modeling and parameter tunning, and     
formulates an optimization objective including both aro-
   
matic yields and energy consumption. Section 3 introduces
the modified DE algorithm in detail, and computational    
results are listed comparing with several distinguished
stochastic optimization algorithms. Section 4 presents the 

results that the proposed algorithm used to solve opti-

mization problem for the CCR process, and the optimal 
operating points are discussed. In Section 5, this article is 
concluded and some lines of future research are mentioned.
Fig. 2. 27-lumped reaction network of catalytic naphtha
2. MODELING OF THE CCR PROCESS reforming

The CCR process is shown in Fig.1. In this process, Naph- sis and paraffin hydro-cracking are regarded as irre-
tha is used as catalytic reforming process feed-stock. It versible reactions.
usually contains more than 300 chemical compounds. The Considering above assumptions, the reaction rate equa-
feed-stock goes through the catalytic reformers combined tions can be established as follow
with a recycle gas stream containing high purity hydrogen.
Catalytic reformer is designed as four cascade reactors. paraffin dehydrocyclization (reversible)
Each reformer is equipped with a heater to maintain the Y5N
ri = ki (Yp − ) (1)
reaction temperature at design level. The effluent from the Kepi
last reactor is cooled and sent to the product separator. naphthene isomerization (reversible)
The separated hydrogen is sent back to reformers as recy-
cle gas, while liquid product is sent to the distillation unit Y6N
ri = ki (Y5N − ) (2)
to obtain aromatic products. Kepi
naphthene dehydroaromatization (reversible)
2.1 Lumps kinetic model YA
ri = ki (Y6N − ) (3)
Kepi
As mentioned above, the naphtha feed-stock is of high
complexity. For this reason, a detailed kinetic reaction paraffin hydrocracking (irreversible)
model will be too complicated to be established. In order YA
ri = ki (Y6N − ) (4)
to reduce the complexity of the kinetic model, components Kepi
with similar reaction performance will be grouped together
as a smaller set of kinetic lumps. Where, ki is reaction rate constant, which can be expressed
as
The lump theory is firstly proposed by Smith (1959). It −Ei
classified naphtha feed-stock as paraffin, naphthene and ki = koi · exp( ) · Phbi · ϕi (5)
RT
aromatic, and contained 4 reactions. Based on this theory,
amount of researches have been taken in decades. Jenkins Where, ϕi , 0 < ϕi < 1 denotes the catalyst active factor.
and Stephens (1980) proposed 31 lumps kinetic model, Radial-flow reactors are adopted as catalytic reformers
Froment (1987) developed a 28 lumps kinetic model. In in the CCR process. The modeling of the radial-flow
2000th, Hongjun et al. (2010) divided the reaction feed- reactors is usually assumed that under the normal reformer
stock and products into 27 lumps, and evaluated the operating conditions, radial and axial dispersion effects are
parameters in a method of fractional steps. negligible, Iranshahi et al. (2010), Gyngazova et al. (2011).
In this work, A 27-lumped kinetic model proposed by The global material and the heat balance equations are
Hongjun et al. (2010) is adopted. The reaction network is given as follow
shown in Fig.2. The network takes the following assump- dY 2πR · H
tion into consideration = · Kr · Y (6)
dR (LHSV · Vc )
• Paraffin cant translate into C6 naphthene and aro- dT 2πR · H (rj · ∆Hj )
= ·Σ (7)
matic directly. dR (LHSV · Vc ) (Cp · Y )
• Neglect the reaction that isoparaffin reacts to form
C6 naphthene. Where, Y is the vector of the molar flow rates including
• Aromatic is of stability, for this reason, aromatic 27 lumps and hydrogen. LHSV denotes the short of liquid
hydro-cracking reaction only happens on side chain. hourly space velocity.
• The rate of naphthene isomerization and dehydro-
cyclization is extremely fast. In consequence, the 2.2 Parameter tunning
cracking reaction among naphthene can be neglected.
• Defining isomerization, dehydrocyclization, and arom- In order to adjust the process model to reflect the actual
atization as reversible reaction, while hydrogenoly- the CCR process accurately, 10 sets of actual operating

374
 IFAC ADCHEM 2015
June 7-10, 2015. Whistler, BC, Canada Min Wei et al. / IFAC-PapersOnLine 48-8 (2015) 373–378 375

Table 1. Operating data from the actual CCR 2.3 Optimization problem for the CCR process
process
In the CCR process, aromatic yields are important as-
F AP T1 T2 T1 T4 p AY
(t/h) (%) (◦ C) (◦ C) (◦ C) (◦ C) (M pag)(%)
sessment standard of the production quality. Meanwhile,
set 1 198.33 42.43 526.99 526.81 524.98 529.02 1.02 64.02 in order to form more C8− aromatic, the yield of heavy
set 2 198.37 39.98 527.04 527.11 525.08 528.86 1.02 62.48 aromatic should be limited. Considering the fact that the
set 3 198.34 43.85 526.99 527.02 524.99 529.03 1.02 64.27 price of energy is rising and low-carbon is a world-wide
set 4 198.38 41.18 527.01 527.00 524.71 529.03 1.02 63.54
set 5 198.27 40.85 526.97 527.00 525.00 528.99 1.02 63.54
issue. In this article, energy consumption is also taken into
set 6 198.33 40.34 526.99 527.00 525.05 528.99 1.02 63.21 consideration.
set 7 198.30 40.10 525.99 525.97 524.00 527.97 1.02 63.92
set 8 198.31 40.08 526.01 525.96 524.04 527.99 1.02 62.75 As mentioned above, this work considers three optimiza-
set 9 198.31 40.10 526.00 525.98 523.95 528.05 1.02 62.50 tion factors, which are maximizing aromatic yields, mini-
set 10 198.30 40.13 525.88 526.14 523.88 527.97 1.02 61.69 mizing heavy aromatic yields, and minimizing energy con-
sumption. Due to the fact that the heat duty from heaters
68 is the uppermost energy consumption in the CCR process.
Adjusted Model In this work, energy consumption is expressed as the fired
66 Original Model duty of the four heaters.
of aromatics yield (m/m)
Model predicted value

64
The function of optimization problem could be formulated
62 as follow
60
max f (x) = ω1 AY + ω2 HAY − ω3 Q (8)

58 where, AY indicates aromatic yield and HAY indicates

heavy aromatic yields. Q denotes the fired duty in heaters.
56
ωi , i = 1, 2, 3 are weight values. These values are all
61 61.5 62 62.5 63 63.5 64 64.5 65 65.5 66
Industrial data of aromatics yield (m/m)
determined by the actual price of the relevant variable. The
decision variable x selected are reactors inlet temperature
(T 1, T 2, T 3 and T 4), reaction pressure (p), and hydrogen-
Fig. 3. Comparison between the adjusted kinetic model
to-oil molar ratio (nH2/nHC). According to the actual
and the original model on the values of aromatic yields
operation data, the constraint of the decision are presented
Table 2. Results of model predict and plant as
actual data (record main components among 510  T1 , T2 , T3 , T4  540 (9)
set 1 only) 0.9  p  1.1 (10)
2.5  nH2 /nHC  540 (11)
composition actual predict dif f erence
NP 4 5.09% 5.36% −0.27%
NP 5 1.42% 0.00% 1.42% 3. DISTANCE BASED DE
SBP 6 3.12% 1.88% 1.23%
NP 6 1.66% 0.66% 0.99% Since an accurately modified CCR model and an opti-
A6 8.07% 7.11% −0.04%
A7 20.25% 20.31% −0.06%
mization problem have been established in section 2. In
EB 5.37% 5.39% −0.02% this section, a modified differential evolution algorithm is
OX 7.88% 7.35% 0.53% proposed to solve this problem.
MX 11.78% 11.70% 0.08%
PX 5.61% 5.47% 0.14% 3.1 Differential evolution
A9 20.50% 20.72% −0.22%

data collected from a certain catalytic reforming unit is The differential evolution (DE) is a population-based s-
used to estimate the reaction parameters. The values of the tochastic optimization algorithm, Price (1996). The per-
data are listed in Table.1. Where, F is the mass flow rate formance of DE is distinguished, especially in convergence
of feedstock. AP denotes the aromatic potential content. speed and robustness.
T1 T4 are the values of inlet temperature in each reactor, In the DE algorithm, mutation and crossover are utilized
p denotes the reaction pressure, and AY is the aromatic to generate trial vectors. For the former operation, the
yields of the products. classic mutation strategy is presented as follow
In this article, the sum of the squared differences of all vit = xtr1 + F (xtr2 − xtr3 ) (12)
components in products is taken as the objective function.
Therefore, the parameter estimation problem is converted Where ri (i = 1, 2, 3) indicates mutually integers selected
into an optimization problem. DE is applied to solve randomly within [1, N P ], and the value of each ri must be
this problem. The difference between the adjusted kinetic unique. xtri indicates individual selected from the popula-
model and the original model is presented in Fig.3. The tion at generation t, and xb estt is the best vector among
data reflect the products aromatic yields of each set. the population. F denotes the scale factor predetermined
mutually. The value of F is usually chosen within (0, 1).
Table.2 shows the detail result of the parameter tunning. It
can be observed that the difference value of main aromatic The strategy that accomplishes the crossover operator is
components are all below 1%. Thus, it could be concluded shown as follow { t
that the adjusted CCR model is able to reflect the special t vi , if (rand  CR)
ui,j = (13)
actual CCR unit accurately. xti,j , otherwise

375
IFAC ADCHEM 2015
376
June 7-10, 2015. Whistler, BC, Canada Min Wei et al. / IFAC-PapersOnLine 48-8 (2015) 373–378

 

 



 



 
 



 
 
                     
 

Fig. 4. The illustration of population distance
Where uti,j indicates j-th element of i-th new trial vector
at generation t.
Finally, offspring can be generated in the use of selection
operation which is presented as follows
{ t
ui , if (f (uti )  f (xti ))
xt+1 = (14)
i xti , otherwise
3.2 Modified DE based on population distance
Due to the advantage of high convergence and robustness,
the DE algorithm is widely used in solving optimization
problems. In decades, a number of modified strategies have
been proposed to develop the performance of DE, some of
which are of high effectivity, such as Fan and Lampinen
(2003), Brest et al. (2006), Rahnamayan et al. (2008), and
Qin et al. (2009). However, most of these researches mainly
focused on the discovery of novel mutation strategies or
the setting of operating parameters, such as F and CR
mentioned above. The importance of population structure
at each generation is always being neglected.
Fig.4 shows a case that the DE algorithm solves a multi-
modal optimization problem with 2-dimensional variables
at a certain generation. It can be observed that individuals
in current population are mainly divided into 4 sets
naturally. In fact, each set surrounds a local optimal. It
can be found that most individuals located in set 1, and
this set also contains the best vector. However, the global
optimum is near set 4.
In this phenomenon, there is high probability that all
xtri would be chosen in set 2. According to the mutation
strategy 12 listed above, if xtri are similar with each other,
the trial vector could not be much different from the
current individuals. That might lead to the deficiencies
of population diversity and too much repetitive counting.
For the reason that calculating the CCR process model is
time consuming, duplicated calculation is unacceptable in
the process optimization.
In order to solve this problem, a modified DE algorithm
based on distance information (DisDE) is proposed in this
work. In this method, individuals at each generation are
distributed into Sn sets by the Euclidean distance from
individuals to the current best vector. Sn is the number
of sets. The equation of counting the Euclidean distance
is presented as follow
Fig. 5. The illustration of the main idea of the DisDE
algorithm
�
�∑
� n 2
di = �
t
(xti,best − xti,j ) (15)
j=1
Where dti
denotes the distance from i-th vector to the
current best vector at generation t. n is dimensionality of
variables. According to the distance value of each vector,
population could be divided into Sn sets by cluster algo-
rithm, where Sn is an operating parameter predetermined.
Thus, the mean distance of each set could be calculated as
∑ dtset
d¯tsetk = k
(16)
Nsetk
Where probtset indicates the number of vectors that belong
to set k, k = 1, ..., Sn . In the use of mean distance value,
selection probability could be obtained as
d¯t
probtsetk = N setk (17)
∑ ¯t
dseti
i=1
Where probtsetk is the probability that trial vector is chosen
from set k. it is obvious that a set with larger distance
value from the best vector might obtain higher probability
to be chosen. In this method, the trial vectors chosen to
generate offspring would be more different from current
individuals. the diversity of population could be kept. The
illustration of the main idea of the DisDE algorithm is
shown in Fig.5. The flowchart of the proposed algorithm
is presented as Fig.6.
3.3 simulation and analysis
In order to demonstrate the performance of the proposed
algorithm, DisDE is calculated by solving 6 benchmark
functions from CEC 2005. These functions are listed in
Table.3. Where D is variable dimensionality, S denotes the
range of variables, and fmin is the global optimal value of
each function. Among the 6 functions, F1 to F4 are single-
modal functions,which are used to test the convergence
speed of algorithms. While F5 and F6 are multi-modal
functions, and these functions are applied to assess the
global search ability.
As comparison, the traditional DE algorithm adopting
DE/rand/1 and the TDE, Fan and Lampinen (2003), are

376
 IFAC ADCHEM 2015
June 7-10, 2015. Whistler, BC, Canada Min Wei et al. / IFAC-PapersOnLine 48-8 (2015) 373–378 377

 Table 4. Results of 6 benchmark functions

DisDE DE T DE
min 7.25E − 18 2.53E − 15 1.88E-22
   F1 (x) mean 1.19E − 17 2.28E − 14 5.64E-21
std 1.90E − 18 1.71E − 14 1.23E-20
min 2.79E-09 2.35E − 07 3.76E − 08
   F2 (x) mean 4.98E-08 7.78E − 07 2.10E − 07
    std 6.60E-08 4.33E − 07 2.83E − 07
min 4.85E − 05 7.49E − 05 1.15E-07
F3 (x) mean 1.30E − 04 2.03E − 04 4.72E-07
   std 1.75E − 04 1.71E − 04 5.36E-06
    min 3.14E-01 2.58E + 01 1.30E + 00
  F4 (x) mean 3.86E-01 3.84E + 01 6.24E + 00
std 7.17E-02 2.64E + 01 2.69E + 00

min 0 0 0
    F5 (x) mean 200 300 300
prob 0.8 0.63 0.5
min 0 0 200
      F6 (x) mean 200 300 300
 
  prob 0.73 0.27 0.10

Table 5. Price of aromatics

 
Ben Tol A8 A9 A10
$/t 1548 1389 1683 1270 1222
mass frac 0.11 0.37 0.52 0.58 0.42
  
It can be observed from the results of F5 and F6 that
 
DisDE both performs best. Especially in solving F6 , due
to the global optimum is located on the bounds, it is very
difficult for algorithms to discover the global optimum.
 This feature can be reflected from the results. It is obvious


that the probability of DE and TDE locating the global
optimum in F6 is low, the value of which are below 30%.
 While, the proposed algorithm located the global optimum
more in 30 calculation times. Thus, it can be concluded

that the modification strategy that we adopt in the DE
algorithm is able to improve the global search ability and
Fig. 6. Flowchart of modified DE algorithm promote the computation efficiency.

4. OPTIMIZATION OF CCR PROCESS BY DISDE

Table 3. Benchmark functions from CEC 2005
Benchmarkf unction D S fmin As the optimization objective function established in 8,
F1 (x): Shifted Sphere Function 30 [−100, 100]D 0
it can be observed that the function is composed of AY ,
F2 (x): Shifted Schwefel’s Probblem 30 [−100, 100]D 0
1.2 HAY , Q, and their weight values. AY and HAY can be
F3 (x): Shifted Schwefel’s Probblem 30 [−100, 100]D 0 estimated as follow
1.2 with Noise in Fitness (FA6 + FA7 + FA8 )
F4 (x): Schwefel’s Problem 2.6 with 30 [−100, 100]D 0 AY = (18)
Global Optimum on Bounds Finlet
D
F5 (x): Rotated Hybrid Composition 2 [−5, 5] 0 (FA9 + FA10 )
Function HAY = (19)
F6 (x): Rotated Hybrid Composition 2 [−5, 5]D 0 Finlet
Function with Global Optimum on
the Bounds Where FAi , i = 6, 7, 8, 9, 10+ denotes the mass flow rate
of structure of aromatic. Finlet is the mass flow rate
of naphtha feedstock. The value of ω1 , i = 1, 2, 3 is
selected to make experimentations. All the algorithms determined by actual price of the relevant variable, which
are predetermined the same operating parameter and is listed in Table.5.
maximum generation, as F = 0.4, CR = 0.9, Gm = 3000,
In addition to the aromatic price and mass flow rate,
P op = 200, and calculated 30 times for each function.
energy consumption data is also available as
Where Gm denotes the maximum generation and P op
indicates the population size. And M t = 0.15 for TDE, 1kg oil = 41.868M J
M d = 0.2 for DisDE. oil price = $770/t
The results of each algorithm solving the mentioned func- According to all data listed above, values of ω1 , i = 1, 2, 3
tions are listed in Table.4. In each function, the best result can be estimated as
is in bold type. For the functions F1 to F4 , the DisDE
ω1 = 3
performs the bestin F2 and F4 , and the results of DisDE
in F1 and F3 is still much better than the original DE ω2 = 2.5
algorithm. ω3 = 1.8 × 10−10

377
IFAC ADCHEM 2015
378
June 7-10, 2015. Whistler, BC, Canada Min Wei et al. / IFAC-PapersOnLine 48-8 (2015) 373–378

Table 6. Results of optimization for the CCR Brest, J., Greiner, S., Boskovic, B., Mernik, M., and
process Zumer, V. (2006). Self-adapting control parameters in
differential evolution: A comparative study on numerical
actual optimal effect
F (t/h) 198.333
benchmark problems. Evolutionary Computation, IEEE
AP (%) 42.425 Transactions on, 10(6), 646–657.
T1 () 525.882 522.307 −3.575 Das, S. and Suganthan, P.N. (2011). Differential evolution:
T2 () 526.137 522.586 −3.551 a survey of the state-of-the-art. Evolutionary Computa-
T3 () 523.877 525.368 1.491
T4 () 527.967 533.07 5.103
tion, IEEE Transactions on, 15(1), 4–31.
p(M P ag) 1.020 0.962 −0.058 Fan, H.Y. and Lampinen, J. (2003). A trigonometric
nH2 /nHC (mol/mol) 3.07 2.78 −0.29 mutation operation to differential evolution. Journal of
AY (%) 61.68 64.02 2.34 Global Optimization, 27(1), 105–129.
HAY (%) 20.63 17.60 −3.03
Q(kJ/h) 3.16E + 08 3.09E + 08 −7.00E + 06
Froment, G. (1987). The kinetics of complex catalytic
reactions. Chemical engineering science, 42(5), 1073–
The result of solving this objective function problem 1087.
applying the proposed algorithm is shown in Table.6. From Gyngazova, M.S., Kravtsov, A.V., Ivanchina, E.D., Ko-
the optimization, it can be observed that the increasing rolenko, M.V., and Chekantsev, N.V. (2011). Reactor
of T1 , T2 , p, and hydrogen-to-oil molar ratio might lead modeling and simulation of moving-bed catalytic re-
to both decreasing of the aromatic yields and heavy forming process. Chemical Engineering Journal, 176,
aromatic yields. With the temperature increasing, more 134–143.
energy would be consumed in vain. An increasing in T3 Hongjun, Z., Mingliang, S., Huixin, W., Zeji, L., and
could obtain higher aromatic yields, but the effect is slight. Hongbo, J. (2010). Modeling and simulation of moving
Meanwhile, a larger value of T4 could result in significant bed reactor for catalytic naphtha reforming. Petroleum
increase in aromatic yields, while decreasing in heavy science and technology, 28(7), 667–676.
aromatic yields. Despite that higher temperature means Iranshahi, D., Pourazadi, E., Paymooni, K., Bahmanpour,
more energy consumption, in the mean of increasing T4 , A., Rahimpour, M., and Shariati, A. (2010). Modeling
the benefits from higher aromatic yields would be much of an axial flow, spherical packed-bed reactor for naph-
larger. tha reforming process in the presence of the catalyst
deactivation. international journal of hydrogen energy,
In Table.6, actual denotes the normal operating point in 35(23), 12784–12799.
the CCR process, optimal lists the suggestion operating Jenkins, J. and Stephens, T. (1980). Kinetics of cat
points, and effect records the difference value between reforming. Hydrocarbon Processing, 60(11), 163–167.
actual and optimal. It can be observed that the aromatic Price, K.V. (1996). Differential evolution: a fast and simple
yields are increased by 2.34% by optimization. Such im- numerical optimizer. In Fuzzy Information Processing
provement is remarkable. Meanwhile, the heavy aromatic Society, 1996. NAFIPS., 1996 Biennial Conference of
yields decreased by 3.03%, which means more heavy aro- the North American, 524–527. IEEE.
matic is formed A6 to A8. Additionally, the total heat Qin, A.K., Huang, V.L., and Suganthan, P.N. (2009). D-
duty decreased by 7.00E + 06kj/h. The reason is that the ifferential evolution algorithm with strategy adaptation
modified algorithm decreased the temperature which has for global numerical optimization. Evolutionary Com-
slight effort to obtain more aromatic, while increased the putation, IEEE Transactions on, 13(2), 398–417.
effective ones. Based on such result, it can be concluded Rahimpour, M.R., Iranshahi, D., and Bahmanpour, A.M.
that in the way of redistributing the operating points, (2010). Dynamic optimization of a multi-stage spherical,
including heatertemperature, pressure and hydrogen − radial flow reactor for the naphtha reforming process in
to−oilmolarratio, the efficiency of the CCR process could the presence of catalyst deactivation using differential
be promoted, while more energy is economized. evolution (de) method. international journal of hydro-
gen energy, 35(14), 7498–7511.
5. CONCLUSION Rahnamayan, S., Tizhoosh, H.R., and Salama, M.M.
(2008). Opposition-based differential evolution. Evo-
In this article, a modified DE algorithm is proposed. The lutionary Computation, IEEE Transactions on, 12(1),
validity of modification is verified by solving 6 benchmark 64–79.
functions, and the proposed algorithm is applied to the op- Smith, R. (1959). Kinetic analysis of naphtha reforming
timization of the continuous catalytic naphtha reforming with platinum catalyst. Chem. Eng. Prog, 55(6), 76–80.
(CCR) process successfully. Xu, B., Qi, R., Zhong, W., Du, W., and Qian, F. (2013).
Optimization of p-xylene oxidation reaction process
The future work will pay attention focus on the investiga- based on self-adaptive multi-objective differential evolu-
tion of classification method used to distribute individuals, tion. Chemometrics and Intelligent Laboratory Systems,
and attempt to make the individuals distribution more 127, 55–62.
reasonable.

REFERENCES
Angira, R. and Santosh, A. (2007). Optimization of
dynamic systems: A trigonometric differential evolution
approach. Computers & Chemical Engineering, 31(9),
1055–1063.

378