Vous êtes sur la page 1sur 10

Article

pubs.acs.org/IECR

Dynamic Optimization of Lurgi Type Methanol Reactor Using Hybrid


GA-GPS Algorithm: The Optimal Shell Temperature Trajectory and
Carbon Dioxide Utilization
Abdulaziz Alarifi,* Zhefu Liu, Fatih Safa Erenay, Ali Elkamel, and Eric Croiset
University of Waterloo Waterloo, Ontario, Canada N2L 3G1

ABSTRACT: At present, methanol is mostly produced from syngas, derived from natural gas through steam methane reforming
(SMR). In a typical methanol production plant, unreacted syngas is recycled for mixing with natural gas and both used as fuel in
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

the reformer furnace resulting in carbon dioxide (CO2) emissions from the flue gases emitted into the atmosphere. However,
CO2 can be captured and utilized as feedstock within the methanol synthesis process to enhance the productivity and efficiency.
To do so, dynamic optimization approaches to derive the ideal operating conditions for a Lurgi type methanol reactor in the
Downloaded via UNIV OF SOUTH DAKOTA on July 7, 2018 at 17:51:51 (UTC).

presence of catalyst deactivation are proposed to determine the optimal use of recycle ratio of CO2 and shell coolant temperature
without violating any process constraints. In this context, this study proposes a new approach based on a hybrid algorithm
combining genetic algorithm (GA) and generalized pattern search (GPS) derivative-free methodologies to provide a sufficiently
good solution to this dynamic optimization problem. The hybrid GA-GPS algorithm has the advantage of sequentially combining
GA and GPS logics: GA, as the most popular evolutionary algorithm, effectively explore the landscape of the fitness function and
identify promising basins of the search space, whereas GPS efficiently searches existing basins in order to find an approximately
optimal solution. The simulation results showed that implementing the shell temperature trajectory derived by the proposed
approach with 5% recycle ratio of CO2 increased the production of methanol by approximately 2.5% compared to the existing
operating conditions.

1. INTRODUCTION CH4 + CO2 → 2CO + 2H 2 ΔH298K = 247.3 kJ mol−1


Carbon dioxide is the primary greenhouse gas emitted to the (1)
atmosphere through fossil fuels (coal, natural gas, and oil) A challenging problem in commercialization of the dry
consumption for energy supply and transportation. In addition, reforming process is that the reaction condition is more
many industrial processes produce CO2 emission either as a favorable for other side reactions such as methane decom-
result of fossil fuel combustion to create heat and steam and/or position (eq 2) and boudouard equilibrium (eq 3).7
as a byproduct of chemical reactions.1 For instance, steam
methane reforming (SMR) is widely used to produce syngas CH4 → C + 2H 2 ΔH298K = 74.8 kJ mol−1 (2)
from natural gas, where the reforming reaction takes place
inside catalyst filled tubes heated externally by firing fuel in a 2CO → C + CO2 ΔH298K = − 172.5 kJ mol−1 (3)
burner located in the furnace chamber.2 Flue gases are
Coke deposition occurs by the exothermic boudouard reaction
discharged through an exhaust at the top of furnace to a heat and the endothermic cracking of methane, which leads to
recovery system where feed fuel and combustion air are severe catalyst deactivation and the need for premature catalyst
preheated, and steam is generated. In the process of burning change-out.8,9 The CALCOR process uses a special catalyst for
natural gas using air as an oxidant, the flue gas usually consists reacting dry CO2 with natural gas, liquid petroleum gas (LPG),
of mostly nitrogen followed by carbon dioxide with nearly 10− and syngas to produce high-purity carbon monoxide that
25 volume percent of the flue gas. It also contains a small contains less than 0.1% methane and has a H2/CO ratio as low
percentage of some pollutants, such as soot, carbon monoxide, as 0.4.10,11 A combination of steam methane reforming and dry
and nitrogen oxide.3 Consequently, research studies on carbon methane reforming (SMR+DRM) of natural gas is a
dioxide utilization focus on developing novel approaches for competitive process to the conventional SMR process. SMR
mitigating the contributions of CO2 emissions to global +DRM produces the best syngas ratio for production of liquid
warming and ocean acidification by developing beneficial uses fuels with a 2−2.1 H2/CO molar ratio, whereas SMR of natural
of CO2 in areas where geologic storage of CO2 may not be the gas produces syngas with a H2/CO molar ratio of slightly above
best solution.4 The advantage of CO2 fixation over CO2 3.12 Regardless of its advantages, SMR+DRM process has not
disposal is that the production of chemicals with high economic been matured yet and still suffers from coking, severely
value is possible.5
The reaction of carbon dioxide and methane in absence of Received: August 7, 2015
steam (eq 1), known as dry reforming of methane (DRM), to Revised: November 26, 2015
form syngas is the most appealing approach to convert two Accepted: December 17, 2015
inexpensive materials into useful chemical building blocks.6 Published: December 17, 2015

© 2015 American Chemical Society 1164 DOI: 10.1021/acs.iecr.5b02918


Ind. Eng. Chem. Res. 2016, 55, 1164−1173
Industrial & Engineering Chemistry Research Article

Figure 1. Scheme of the methanol synthesis loop.

shortening the lifetime of the applied catalysts. Because the recycle ratio of CO2 within the design limit of the plant
DRM technology has not been widely implemented for (maximum 10%) and the peak bed temperature less than or
reducing CO2 emission, alternative approaches to address this equal to 543 K and then analyzed the effects of these variables
issue are still under ongoing investigation.13 Chemical recycling on the reactor performance and the catalyst activity.
of CO2 to renewable fuel, such as methanol, offers a future
transportation fuel. 2. MODEL DEVELOPMENT
Olah and his research group developed an improved Cu/
ZnO-based catalyst for methanol synthesis from CO2 and H2 Consider a midsize methanol plant, e.g., up to 2500 ton/day,
under operating conditions and deactivation rate similar to that employs a steam methane reformer with an external source
those of commercial catalysts for syngas-based methanol of hot gas to heat tubes in which catalytic reforming reactions
synthesis.14,15 The first CO2-to-methanol recycling plant take place. The feedstock to the reformer is a mixture of steam
(George Olah Renewable Methanol Plant) is operated in and natural gas with a typical steam-to-carbon ratio of 3:1 to
Iceland by Carbon Recycling International (CRI).16 This plant avoid carbon deposition on the catalyst surface.21 Steam and
uses available local geothermal energy (hot water and steam) to methane are converted to syngas at a high temperature and an
supply the electricity needed to produce pure H2 by water elevated pressure. The syngas is then mixed with recycled
electrolysis process with a daily capacity of 10 ton/day. The unreacted gas, pressurized in a compressor until the operational
present technologies of water electrolysis have an efficiency of pressure of methanol synthesis process is achieved, and then
50−80% and require 50−79 KWh/kg of H2; hence, they cost fed to a water-cooled tubular reactor. The tubes packed with
3−10 times more than steam methane reforming even with a catalyst are externally cooled by boiler water, and a nearly
traditional method of generating electricity.17 isothermal temperature profile is formed in the catalyst bed.
The direct conversion of natural gas is more feasible and
This type of reactor aims at recovering the heat of methanol
cost-effective at an industrial level than burning the fuel to
generate electricity needed to electrolyze water and then synthesis reactions to the maximum possible extent in the form
produce hydrogen.18 For an SMR-based methanol plant, the of high-pressure steam. Figure 1 shows the schematic diagram
H2/CO molar ratio obtained is close to 3, which is not optimal of the methanol synthesis process. Methanol synthesis from
for methanol synthesis process. Therefore, recycling CO2 back syngas over Cu/ZnO/Al2O3 catalyst is composed of three main
to the reformer feed gas can be an option to increase methanol equilibrium reactions:
production by adjusting the stoichiometry of the makeup gas,
which increases the reformer heat duty while reducing the CO + 2H 2 ⇌ CH3OH ΔH298K = − 91 kJ mol−1 (4)
amount of wastewater discharged from the distillation system.
A case study presented by Reddy et al. showed that the CO2 + 3H 2 ⇌ CH3OH + H 2O ΔH298K = − 50 kJ mol−1
methanol plant capacity can be expanded approximately 20% (5)
by recovering sufficient CO2 from steam reformer stack and
recycling it as feedstock to the SMR.19 However, recycling CO2
to the reformer feed-gas increases the demand for heat. CO2 + H 2 ⇌ CO + H 2O ΔH298K = + 41 kJ mol−1 (6)
An alternative option is recycling CO2 into the syngas
compressor suction feed rather than reformer feed. In a 2.1. Reactor Model. The methanol synthesis reactor
previous research, we have concluded that around 3% studied here is a Lurgi type used in Shiraz Petrochemical
improvement in the plant production capacity can be achieved Company.22 A 1D heterogeneous reactor model is used to
by injecting additional interstage CO2 into ICI’s multibed simulate the performance of industrial Lurgi type methanol
adiabatic reactor.20 The objective of this study is to use dynamic reactor. It employed a two-phase model in which heat transfer
optimization to derive the optimal recycle ratio of CO2 and through the solid and fluid phases is considered separately.23
shell temperature trajectory of an industrial methanol Lurgi The continuity and energy equations for solid and fluid phases
type reactor during four years of catalyst lifetime. We kept the are expressed as follows:24
1165 DOI: 10.1021/acs.iecr.5b02918
Ind. Eng. Chem. Res. 2016, 55, 1164−1173
Industrial & Engineering Chemistry Research Article

Continuity equations: Table 1. Feed Composition and Catalyst Specifications of


∂C ∂C Industrial Plant Methanol Synthesis Reactor
εb i = −uz i + kg, iav(Cis − Ci)
∂t ∂z parameters value
∀ z ∈ (0, L], t ∈ (0, t f ], i = 1, 2, ..., Nc (7) Feed Composition
Inlet flow rate (mol s−1) 0.64
∂Cis mole fraction %
εs = R i(Cis ,
T)− s
kg, iav(Cis
− Ci) CH3OH 0.50
∂t
CO 4.60
∀ z ∈ (0, L], t ∈ (0, t f ], i = 1, 2, ..., Nc (8) CO2 9.40
Energy equations: H2 65.90
H2O 0.04
∂T ∂T
ρf cp ,f = −uzρf cp ,f + hf av(T s − T ) CH4 10.26
∂t ∂z N2 9.30
U Design and Catalyst Specifications
+ 4 w (Tshell − T ) ∀ z ∈ (0, L], t ∈ (0, t f ] Inlet pressure (bar) 76.98
Di (9)
Length of the reactor (m) 7.022
∂T s εb 0.39
ρs cp ,s = RT(Cis , T s) − hf av(T s − T ) ρb (kg m−3) 1132
∂t
aν (m−1) 626.98
∀ z ∈ (0, L], t ∈ (0, t f ] (10) Tube inner diameter (mm) 38
Number of tubes (Ntubes) 2962
The reaction rates and heat of reactions terms are
Nrxn ⎡ PH OPCO ⎤
R i(Cis , T s) = ρb a ̃ ∑ vijrj(Cis , T s) k 2PCO2⎢1 − Keq2 P 2P ⎥
i = 1, 2, ..., Nc ⎣ H2 CO2 ⎦
j=1 (11) rRWGS = PH O
1 + K ad1 P 2 + K ad2PH0.52 + K ad3PH2O
Nrxn H2 (18)
RT(Cis , T s) = ρb a ̃ ∑ rj(Cis , T s)( −ΔHr, j(T s)) where Pi is the partial pressure of component i in bar, and the
j=1 reaction rates are calculated per kilogram of catalyst (kmol/kg
i = 1, 2, ..., Nc cat s). The parameters of the equations including the reaction
(12)
rate constants, adsorption equilibrium constants, and reaction
Boundary and initial conditions: equilibrium constants are given elsewhere.26
2.3. Deactivation Model. The sulfur components normally
Ci = Ci,in ; T = Tin @z = 0, t ∈ [0, tf ] (13) present in the natural gas feedstock must be removed to
prevent poisons and subsequent quick deactivation of the
Ci = Ci,0 ; T = T0 @t = 0, z ∈ (0, 1] (14) reforming process and methanol synthesis catalysts. Under
normal operation conditions, deactivation of Cu/Zn/Al2O3
Cis = Cis,0 ; T s = T0s @t = 0, z ∈ [0, 1] (15) catalyst is slow and only caused by thermal sintering, which
The methanol production rate can be formulated as occurs by the migration of copper crystals to larger
agglomerates.27 In preparing a fresh catalyst, it is first activated
Production = Ntubes Ac uz M wMeOH C MeOH|z = 1 (16) by reduction conventionally carried out with hydrogen. Once
In this set of equations, Ci stands for concentration of reactant i. the reduction reaction is complete, the fresh catalyst is used
The feed specifications and catalyst characteristics of methanol immediately in methanol synthesis by replacing the hydrogen
synthesis reactor are presented in Table 1. stream by syngas stream and implementing changes in pressure,
2.2. Reaction Kinetics. Three overall reactions (eqs 4−6, temperature, and space velocity, as desired. The fresh catalyst
hydrogenation of CO and CO2, and the reverse water−gas activity declines in the first 20 h by as much as 60%.28 The
shift) are possible occurred in methanol synthesis. Several adjusted deactivation model developed by Hanken was found
studies proposed kinetic models of methanol synthesis. In this to give results in good agreement with industrial operating
study, we used Langmuir−Hinshelwood−Hougen−Watson conditions.29 The rate of change in the activity of Cu/Zn/Al2O3
(LHHW) type reaction kinetic model of Brussche and Froment catalyst based on this model is given as follows:
because it is the only model based on experiments involving a ⎛ −E ⎛ 1
da 1 ⎞⎞ 5
commercial ICI 51−2 Cu/ZnO/Al2O3 catalyst that is grounded = −Kd exp⎜⎜ d ⎜ − ⎟⎟⎟ a ∀ t ∈ (0, t f ]
and diluted for the sake of preventing the diffusion resistance dt ⎝ R ⎝T TR ⎠⎠
and accurately controlling the pseudoisothermicity of the (19)
reaction.25 The corresponding rate expressions for hydro-
a −a
genation of CO and CO2 and reversed water−gas shift ã = 1 − 0
reactions are a0 (20)
⎡ 1 PH2OPCH3OH ⎤ where TR = 513 K, Ed = 91 270 J mol−1, and Kd = 0.00439 h−1
k1PCO2PH2⎢1 − K eq1 PH3 2PCO2 ⎥
⎣ ⎦ are the reference temperature, activation energy, and
rCH3OH = deactivation constant of the catalyst, respectively. An initial
⎡ PH2O 0.5 ⎤3
⎢⎣1 + K ad1 PH2 + K ad2PH2 + K ad3PH2O⎥⎦ value a (t = 0) of 0.4 was assumed. A relative activity started at
(17) 1 is used to find the reaction rate.
1166 DOI: 10.1021/acs.iecr.5b02918
Ind. Eng. Chem. Res. 2016, 55, 1164−1173
Industrial & Engineering Chemistry Research Article

Figure 2. Temperature and concentration profiles calculated by Runge−Kutta fourth-order and explicit finite difference methods.

2.4. Simulation. A mathematical treatment of the du ⃑


= f (z , u ⃑) with u ⃑(0) = u0
governing model, eqs 7−20, is carefully addressed before dz (21)
solving them. Because the catalyst activity rapidly declined
within the first 20 h of start-up operation, the transient behavior where du ⃑ /dz are the derivatives of concentrations and
of the reactor system is not of interest to our investigation, and temperature with respect to space-direction in fluid phase
the numerical solutions are performed in two consecutive (eqs 7 and 9) and the Jacobian matrix of the set of functions f is
stages: steady-state stage followed by dynamic stage. defined as
2.4.1. Steady State. The above heterogeneous model, eqs
7−18, consists of a set of partial differential equations. In ⎡ ∂f ⎤
steady-state model, these equations are reduced to ordinary J f = ⎢ i ⎥ , i = 1, 2, ..., Neqs , j = 1, 2, ..., Nvars
⎢⎣ ∂uj ⎥⎦ (22)
differential equations (ODEs) and algebraic equations by
eliminating the left-hand sides of the equations (unsteady-state The eigenvalues λ are found using the characteristic equation of
terms) and assuming that the catalyst activity is unity. This the square matrix Jf, det(A − λI) = 0. The solution is
system of equations is initially solved by decoupling the fluid conditionally stable if each eigenvalue satisfies the following
and solid phases in such a way that the solution of system is test:
obtained in two consecutive stages, where the solid-phase
nonlinear algebraic equations is solved by using the powerful Re(λi) = |1 + hλi| ≤ 1 (23)
Newton−Raphson method; subsequently, the obtained values
of solid-phase variables are used in the fluid-phase ODEs. The step size (h) is chosen to be relatively small in the solution
ODEs are often solved by Runge−Kutta fourth-order (RK4) or region where the curves display much variation and to be
Gear’s methods. Both methods incorporate an error control relatively large where the solution curves straightness out to
scheme to automatically adjust the step-size, generally approach lines with a slope of almost zero.32 Before we attempt
providing good accuracy. Gear’s method is a more sophisticated to adopt the step-size in solving the dynamic model by using
scheme, ensuring reliable convergence and accuracy; thus, it is method of lines, it is useful to compare the solutions obtained
an effective method for solving a stiff ODEs system.30 by both approaches. It is pointed out that RK4 and FD
Fortunately, the ODEs system considered herein is not stiff; methods give almost identical results for all the key
therefore, a RK4 scheme is applied to provide an accurate components and temperature as shown in Figure 2.
solution with reduced computational costs. The purpose of 2.4.2. Dynamic. As already mentioned earlier, the
solving the system by this technique is to (1) determine the heterogeneous model of tubular reactor comprises approx-
initial conditions for temperature and concentrations to be used imately 15 partial differential and algebraic equations. The
for dynamic simulation described in the next section and (2) to method of lines has been used for solving partial differential
ensure that the integration step size used in the finite difference equations in which a forward finite difference method is used to
method guaranties an acceptable accuracy of the solution and is discretize the spatial domain with nearly 640 finite elements;
good enough to be adopted later to solve the dynamic then, the resulting set of ODEs was numerically integrated
simulation. Considerable attention has been directed toward through time. Dynamic simulator code was developed using eqs
the discretization of the convection terms and incorporation of 7−18, accompanied by catalyst deactivation (eqs 19 and 20) in
source terms, transfer terms, and the boundary conditions into order to compute the reactor effluent and validate the
the numerical solution scheme. The scheme is an explicit mathematical model with the industrial data. Table 2 shows
approximation of the ODEs system, which usually cuts down the predicated production rate of the reactor compared to the
the computational time. However, the implementation of the plant data over an operating time frame of about 1200 days.
explicit method leads to restrictions on the step size because of The dynamic simulation results match well with the industrial
numerical stability concerns. For discretization of the ODEs data; therefore, we concurred that dynamic model with the
system, we imposed the scheme to obey the following stability proposed deactivation model parameters is reliable in
condition:31 predicting the reactor behavior.
1167 DOI: 10.1021/acs.iecr.5b02918
Ind. Eng. Chem. Res. 2016, 55, 1164−1173
Industrial & Engineering Chemistry Research Article

Table 2. Comparison between Plant Data and Predicted because it is restricted by the plant’s design flexibility. There is
Methanol Production only one path constraint; the gas temperature along the reactor
length must be below 543 K to prevent a severe deactivation
time (day) plant (ton/day) prediction (ton/day) % error
rate. The objective function J is the value of the total methanol
0 295 292 −1.01 production over the catalyst lifetime period. The proposed
100 296.5 286.6 −3.33 hybrid algorithm was used for dynamic constrained optimiza-
200 302.6 283.2 −6.41 tion of the reactor operating conditions considering the above
300 284.3 280.6 −1.30 inequality constraints.
400 277.9 278.4 0.17 2.6. Hybrid Genetic-Generalized Pattern Search
500 278.2 276.7 −0.53 Algorithm (GA-GPS) for the Optimization Problem.
600 253.0 275.0 8.69 Genetic algorithm (GA) is the most popular type of
700 274.0 273.4 −0.21 evolutionary algorithms that mimics the principle of biological
800 268.1 272.1 1.49 evolution, repeatedly modifying a sample solution population
900 275.5 270.8 −1.70 using a technique inspired by natural evolution using random
1000 274.6 269.7 −1.78 genetic operators such as mutation, selection, and crossover.33
1100 262.9 268.6 2.16 GA is widely classified as a metaheuristic optimization
1200 255.2 267.6 4.85 algorithm because of its stochastic search mechanism that
enables it to escape from local optima while exploring the
Figure 3 shows the temperature profile and deactivation rate search space for the global optimum in complex and rugged
of catalyst with respect to the reactor length and the time fitness landscapes. 34 In any metaheuristic algorithm, a
horizon according to the dynamic stage simulation results. The compromise between exploration and exploitation components
catalyst activity drops to the lowest level (<0.5 at the end of is carried out to balance the convergence rate of the algorithm
run). Consequently, the maximum reactor temperature is 534 and the likelihood of achieving the global optimality.35,36
K at the start of run and gradually decreases over time. Exploration aims to generate diverse solutions by exploring a
2.5. Problem Formulation. The task is finding the optimal large solution space, whereas exploitation narrows down the
operation strategy, namely, the shell coolant temperature search to a specific regions with high likelihood of containing a
trajectory and recycle ratio of CO2 (RCO2) fed to the reactor. good solution. During the initialization step of GA, an initial
The objective was formulated as a nonlinear dynamic population consisting of a number of individuals (NI) (each
optimization problem as follows: having a set of chromosomes or genotypes (NG) to represent
the decision variables for a particular solution) is randomly
⎡ tf ⎤ generated from the feasible search space. A fitness function is
max⎢J =
⎣ ∫0 Production(R CO2 , Tshell , t ) dt ⎥
⎦ (24) evaluated for each individual; then, the population is sorted by
descending order of the fitness values. Subsequently, the
subject to selection process considers only the fitness values regardless of
0 ≤ T ≤ 543K (25) the individuals’ genotypes. Therefore, the selection is an
exploitative operation because the mechanism does not care
0 ≤ R CO2 ≤ 0.1 (26) about the diversity of genotypes of each selected individuals but
490K ≤ Tshell ≤ 540K (27)
rather distinguishes them on the basis of their fitness values.
These selected individuals are used to perform the genetic
The production rate function is obtained by solving the partial operations: (1) mutating some individuals by replacing some of
differential equations (eq 7−20). The shell coolant temperature the existing genotypes with new genotypes. (2) Crossover
is bounded between 490 and 540 K, and the recycle ratio of (exchanging a portion of genotype) is performed between two
CO2, defined as RCO2 = FCO2/Fin, must be below or equal 0.1 individuals (parents) to create two new individuals (children).

Figure 3. Temperature and catalyst activity profiles for the reference case condition (RCO2 = 0, Tc = 524 K).

1168 DOI: 10.1021/acs.iecr.5b02918


Ind. Eng. Chem. Res. 2016, 55, 1164−1173
Industrial & Engineering Chemistry Research Article

Figure 4. Best and mean fitness values and the average distance between individuals.

Figure 5. Flowchart of hybrid GA-GPS algorithm.

Mutation and crossover are explorative operators because they GA has been proven to be effective in escaping local optima
introduce new genotypes to the population and spread the new and finding solutions closed to the global optimum/optima for
generation over the search space with the hope of exploring even a very complex function. However, it shows weakness in
new basins.33 After enough generations, existing genotypes in efficiently exploiting the existing basins in order to find the
the population become similar to each other, and the mean optimal point.37 Hence, it could be much efficient to use the
fitness value of the populations becomes close to the best output of GA as an input for a local search algorithm such as
fitness value as illustrated in Figure 4. generalized pattern search (GPS). GPS is a derivative-free
1169 DOI: 10.1021/acs.iecr.5b02918
Ind. Eng. Chem. Res. 2016, 55, 1164−1173
Industrial & Engineering Chemistry Research Article

Figure 6. Predicted methanol production rate at different recycle ratio of CO2 ratio associated with the optimal shell temperature trajectory.

optimization method that does not use the information about adjoining segment of a line in such a way that each individual’s
gradient or higher derivative to search for an optimal point.38 segment is equal in size to its scores. SUS generates a single
Starting from an initial solution x0 and an initial mesh size Δm0 = random value (specifying the starting point in [0, 1/Np] to
1, GPS generates a set of neighborhood points by multiplying sample all of the solutions by selecting an individual at every 1/
the current mesh size by each pattern vector. These vectors are Np distance where Np is the number of individuals to select.
fixed-direction vectors and are defined by the number of SUS shows no bias, giving the low-scored individuals a chance
independent variables in the objective function; as in GPS, 2N to be chosen. The next step is to generate the second
vectors consist of N positive and N negative vectors. In k generation of population by applying mutation and crossover
iteration, GPS algorithm polls all the mesh points by computing on the selected individuals, where 80% of the next population is
their objective function values in order to find an improved produced by using GA’s operators (crossover and mutation)
point. If the polling step fails to find an improved point, then and 20% elite individuals from the current and previous
the mesh size will be reduced by half, Δmi+1 = 0.5Δmi , and the generations. The second phase involves an exploitative search
current point is used for the next iteration k + 1. Otherwise, employing GPS after receiving the best solution from GA and
when a successful polling takes place and an improved point is using it as an initial solution for GPS. A complete polling occurs
found, then the mesh size will be doubled, Δmi+1 = 2Δmi , and the by computing the objective function values at all mesh points
current point is updated by the improved point for the next f(Xmi ) starting at the first iteration Δm0 = 1 until some
iteration as shown in Figure 5. termination criteria are met (either Δmi = 10−6 or 300
The incorporation of a local search algorithm such as GPS is iterations).
a good strategy for improving the accuracy of the approximate
solution previously provided by the genetic algorithm. Thus, an 3. RESULTS AND DISCUSSION
appropriate sequential hybridization of GA and GPS can The time horizon for the decision variables of the considered
improve the exploitation of the basin of solution that was optimization problem, shell temperature and recycle ratio of
already explored by GA and can lead to find, as accurately as CO2, were discretized into 16 intervals, each interval is a
possible, a local optimum. To maximize methanol production, quarter of a year over the catalyst lifetime (four years). The
we implemented such a GA-GPS hybrid algorithm to solve the optimization of reactor was performed in two steps: dynamic
problem of optimizing aforementioned reactor’s operation stage optimization of shell temperature followed by simulta-
conditions. neous dynamic stage optimization of shell temperature and
The first phase of GA-GPS hybrid algorithm is exploration recycle ratio of CO2.
via employing GA to minimize the fitness function, negative of 3.1. Dynamic Staged Optimization of Shell Temper-
the total production over the catalyst lifetime (−J). GA ature. For this section, the ratio of recycled CO2 directly fed to
parameters were tuned by troubleshooting, and the number of the reactor is set as 0, 5, and 10% of the total flow rate of the
generations was set to 450. As indicated in Figure 4, the mean inlet stream; consequently, eq 26 is ignored. Figure 6 shows the
value of the fitness function is matched up to the best fitness amount of methanol production rate associated with the
value and no more diversity of individuals. Each generation has approximately optimal shell temperature trajectories derived by
a default population of 200 individuals, and the algorithm the proposed GA-GPS algorithm. The proposed Tshell trajectory
begins by creating a random initial population. These improved the associated production rates for 0, 5, and 10%
individuals represent shell temperature and recycle ratio of recycle ratio of CO2 by 1.67, 2.53, and 1.43%, respectively,
CO2 at each time segment. Then, the fitness value (score) of compared to the reference case of keeping the shell
each individual is evaluated to select the best individuals for temperature constantly at 524 K. The optimal shell temperature
mutation and crossover processes. We used stochastic universal trajectories for these cases gradually increase with time as
sampling (SUS) for the selection operation, where all expected to compensate for catalyst deactivation. In addition,
individuals’ scores are normalized and then mapped to an the more CO2 is fed to the system, the higher Tshell is needed to
1170 DOI: 10.1021/acs.iecr.5b02918
Ind. Eng. Chem. Res. 2016, 55, 1164−1173
Industrial & Engineering Chemistry Research Article

Figure 7. Reactor temperature and the average catalyst activity profiles at the optimal case (the optimal case: 5% recycle ratio of CO2 and shell
temperature trajectory shown in Figure 6).

Figure 8. Predicted methanol production rates associated with the optimal shell temperature and recycle ratio of CO2 trajectories.

supply more heat to the endothermic reaction (eq 6). Figure 6 time horizon for operational simplicity without significantly
shows that the optimal solution does not violate the path reducing the production rate. This way, the number of variables
temperature constraint (eq 27). can be reduced for this optimization problem from 32 to 16 by
The average catalyst activity versus time from both the assuming a fixed amount of recycle ratio of CO2 is utilized (5%)
reference and optimal cases is shown in Figure 7. The average to obtain the optimal production rate.
catalyst activity in the optimal case is higher than in the
reference case as the reactor is operated at lower shell 4. CONCLUSIONS
temperature during the whole lifetime cycle of the catalyst An industrial Lurgi Type Methanol Reactor reactor with
except the last two quarters. catalyst deactivation has been modeled and optimized with
3.2. Simultaneous Dynamic Stage Optimization of respect to the shell coolant temperature and the recycle ratio of
Shell Temperature and Recycle Ratio of CO2. For this CO2 fed to the system. The results showed that correcting the
section, we simultaneously optimized shell coolant temperature syngas ratio of the inlet stream of the reactor system improves
and recycle ratio of CO2 to maximize the production rate. the production rate by increasing CO2 recycled ratio to 5% of
Figure 8 compares the production rates and Tshell trajectories recovered CO2 from the reformer flue gases. The production
for the proposed solution of the simultaneous optimization rates for 0 and 5% recycled CO2 ratio cases were improved by
(referred to as 0−10% CO2) with those for the proposed 1.67 and 2.53%, respectively. These improvements are
solution with fixed 5% recycle ratio of CO2. As shown in Figure equivalent to 2.73 and 4.13 million U.S. dollars based on an
8, the approximately optimal production rate for the 5% CO2 average price of 400 per ton of methanol, respectively, over four
and 0−10% CO2 are remarkably close to each other. This result years of operation compared to the reference case (Tshell = 524
implies that the recycle ratio of CO2 can be kept at 5% over the K and no extra CO2 added to the system). Because no tax is
1171 DOI: 10.1021/acs.iecr.5b02918
Ind. Eng. Chem. Res. 2016, 55, 1164−1173
Industrial & Engineering Chemistry Research Article

collected on fossil fuel production in the region where the plant RWGS = Reverse water gas shift reaction.


is located (Middle East), this benefit calculation excluded the
additional profit from avoiding/reducing the charge/tax for REFERENCES
CO2 emission into the atmosphere.


(1) Aresta, M.; Quaranta, E.; Tommasi, I. Carbon Dioxide Utilisation
AUTHOR INFORMATION in the Chemical Industry. Prepr. Pap.-Am. Chem. Soc., Div. Fuel Chem.
1996, 41, 1341.
Corresponding Author (2) Gunardson, H. Industrial Gases in Petrochemical Processing; Marcel
*Tel.: 519-888-4567, ext. 37157. E-mail: aalarifi@uwaterloo.ca. Dekker Inc.: New York, 1997.
Notes (3) Beychok, M. Fossil Fuel Combustion Flue Gases. Encyclopedia of
The authors declare no competing financial interest. Earth, 2012. http://www.eoearth.org/view/article/


51cbf2487896bb431f6a896e/ (Accessed on [8/1/2015]).
NOMENCLATURE (4) Metz, B.; Davidson, O.; de Coninck, H.; Loos, M.; Meyer, L.
IPCC Special Report on Carbon Dioxide Capture and Storage;
Parameters Cambridge University Press: Cambridge, U.K., 2005.
a = Activity of catalyst. (5) Treacy, D.; Ross, J. The Potential of the CO2 Reforming of CH4
Ac = Cross-sectional area of the reactor, m2. as a Method of CO2 Mitigation. A Thermodynamic Study. Prepr. Pap.-
aν = Specific surface area of catalyst pellet, m2 m−3. Am. Chem. Soc., Div. Fuel Chem. 2004, 49, 126.
Ci, Csi = Concentration of component i in gas and catalyst (6) Halmann, M. Chemical Fixation of Carbon Dioxide: Methods for
Recycling CO2 into Useful Products; CRC Press: Boca Raton, FL, 1993.
surface, mol m−3. (7) Suib, S., Ed. New and Future Developments in Catalysis: Activation
Cp, Cp,s = Heat capacity of gas and catalyst particle, kJ kmol−1 of Carbon Dioxide; Elsevier: Amsterdam, The Netherlands, 2013.
K−1. (8) Jessop, P. Greenhouse Gas Carbon Dioxide Mitigation: Science
Di = Internal diameter of tubes, m. and Technology By Martin M. Halmann (Weizmann Institute of
Ed = Activation energy used in the deactivation model, J Science, Israel) and Meyer Steinberg. J. Am. Chem. Soc. 2001, 123,
mol−1. 7197.
Fi = Molar flow rate per tube, mol s−1. (9) Halmann, M.; Steinberg, M. Greenhouse Gas Carbon Dioxide
hf = Fluid−solid heat transfer coefficient, Wm2− K−1. Mitigation: Science and Technology; Lewis: Boca Raton, FL, 1998.
J = Total production, tonne. (10) Teuner, ST. C.; Neumann, P.; Von Linde, F. The Calcor
Kad,j = Adsorption equilibrium constant of component j. Standard and Calcor Economy Processes. Oil Gas Eur. Mag. 2001, 3,
Kd = Deactivation model parameter constant, s−1. 44.
(11) Kurz, G.; Teuner, S. Calcor Process for CO Production. Erdöl,
Keq,i = Equilibrium constant of reaction i. Kohle, Erdgas, Petrochemie 1990, 43, 171.
kg,i = Mass transfer coefficient for component i, m s−1. (12) Chemsystems PERP Program. Methanol; Nexant Report 07/08-
ki = Rate constant of reaction i. 2; 2008.
L = Length of reactor, m. (13) Shi, F., Ed. Reactor and Process Design in Sustainable Energy
Mw MeOH = Molecular weight of methanol, g/mol. Technology; Elsevier: Amsterdam, The Netherlands, 2014
Neqs = Number of equations. (14) Olah, G. a. Beyond Oil and Gas: The Methanol Economy.
Nc = Number of components. Angew. Chem., Int. Ed. 2005, 44, 2636.
Nrxn = Number of reaction. (15) Olah, G. a; Goeppert, A.; Prakash, G. K. S. Chemical Recycling
Ntubes = Number of tubes. of Carbon Dioxide to Methanol and Dimethyl Ether: From
Nvars = Number of variables. Greenhouse Gas to Renewable, Environmentally Carbon Neutral
Pj = Partial pressure of component j, bar. Fuels and Synthetic Hydrocarbons. J. Org. Chem. 2009, 74, 487.
(16) Olah, G.; Prakash, G. K. S. Efficient and Selective Conversion of
rCH3OH = Rate of methanol synthesis reaction, kmol hr−1 kg Carbon Dioxide to Methanol, Dimethyl Ether, and Derived Products.
cat−1. U.S. Patent No. US7605293 B2, October 20, 2009.
ri = Rate of reaction i, mol kg−1 s−1. (17) Zittel, W.; Wurster, R. Hydrogen in the Energy Sector;
Ri = Rate of reaction for component i, mol kg−1 s−1. Hydrogen and Fuel Cell News, Issue 8.7.1996; Ludwig-Bö lkow-
rRWGS = Rate of water gas shift reaction, kmol hr−1 kg cat−1. Systemtechnik GmbH, 1996.
RT = Heat generated from all the reactions, kJ. (18) Badwal, S. P. S.; Giddey, S. S.; Munnings, C.; Bhatt, A. I;
T = Gas phase temperature, K. Hollenkamp, A. F. Emerging Electrochemical Energy Conversion and
t = Time, s. Storage Technologies. Front. Chem. 2014, 2, 79 DOI: 10.3389/
TR = Reference temperature, K. fchem.2014.00079.
Ts = Catalyst temperature, K. (19) Reddy, S.; Bhakta, M.; Gilmartin, J.; Yonkoski, J. Cost Effective
Tshell = Shell coolant temperature, K. CO2 Capture from Flue Gas for Increasing Methanol Plant
Production. Energy Procedia 2014, 63, 1407.
uz = Gas velocity in axial axis, m s−1.
(20) Alarifi, A.; Alsobhi, S.; Elkamel, A.; Croiset, E. Multiobjective
Uw = Overall heat transfer coefficient, Wm2− K−1. Optimization of Methanol Synthesis Loop from Synthesis Gas via a
z = Axial direction of the reactor. Multibed Adiabatic Reactor with Additional Interstage CO2 Quench-
Greek Letters and Symbols ing. Energy Fuels 2015, 29, 530.
ΔHr,j = Heat of reaction j, kJ kmol−1. (21) Filippi, E. Process for Producing Methanol from Steam
νi,j = Stoichiometric coefficient of component i in reaction j. Reforming. WO Patent No. 2,009,127,657, 2009.
ηn = Effectiveness factor. (22) SPC. Methanol Reactor Long Sheet; Shiraz Petrochemical
ρb, ρs = Bulk and solid density of catalyst, kg m−3. Company: 2000−2003.
(23) Koning, B. Heat and Mass Transport in Tubular Packed Bed
ρf = Molar density of gas, mol m−3.
Reactors at Reacting and Non-Reacting Conditions, Ph.D. Thesis,
εb, εs = Void fraction in catalyst bed and catalyst particle. University of Twente, Enschede, The Netherlands, 2002.
Abbreviations (24) Froment, G. F.; Bischoff, K. B. Chemical Reactor Analysis and
MTPD = Metric ton per day. Design, 2nd ed.; Wiley: New York, 1979.

1172 DOI: 10.1021/acs.iecr.5b02918


Ind. Eng. Chem. Res. 2016, 55, 1164−1173
Industrial & Engineering Chemistry Research Article

(25) Bussche, K. M. V.; Froment, G. F. A Steady-State Kinetic Model


for Methanol Synthesis and the Water Gas Shift Reaction on a
Commercial Cu/ZnO/Al2O3 Catalyst. J. Catal. 1996, 161, 1.
(26) Alarifi, A.; Elkamel, A.; Croiset, E. Steady-State Simulation of a
Novel Annular Multitubular Reactor for Enhanced Methanol
Production. Ind. Eng. Chem. Res. 2013, 52, 15387.
(27) Forzatti, P.; Lietti, L. Catalyst Deactivation. Catal. Today 1999,
52, 165.
(28) Løvik, I.; Hillestad, M.; Hertzberg, T. Modeling and
Optimization of a Reactor System with Deactivating Catalyst. Comput.
Chem. Eng. 1999, 23, S839.
(29) Hanken, L. Optimization of Methanol Reactor. Master Thesis,
Norwegian University of Science and Technology, 1995.
(30) Hartman, P. Ordinary Differential Equations; Wiley: New York,
1964.
(31) Heath, M. T. Scientific Computing: An Introductory Survey;
McGraw-Hill: Boston, 1997
(32) Liu, S. Stable Explicit Difference Approximations to Parabolic
Partial Differential Equations. AIChE J. 1969, 15, 334.
(33) Goldberg, D. E. Genetic Algorithms in Search, Optimization, and
Machine Learning; Addison-Wesley: Reading, MA, 1989.
(34) Bajpai, P.; Kumar, M. Genetic Algorithm−an Approach to Solve
Global Optimization Problems. Indian J. Comput. Sci. Eng. 2010, 1,
199.
(35) Holland, J. H. Adaptation in Natural and Artificial Systems;
University of Michigan Press : Ann Arbor, MI, 1975.
(36) Blum, C.; Roli, A. Metaheuristics in Combinatorial Optimiza-
tion: Overview and Conceptual Comparison. ACM Comput. Surv.
2003, 35, 268.
(37) Preux, P.; Talbi, E. G. Towards Hybrid Evolutionary Algorithms.
Int. Trans. Oper. Res. 1999, 6, 557.
(38) Rios, L. Algorithms for Derivative-Free Optimization, Ph.D.
Thesis, Department of Industrial and Enterprise Systems Engineering,
University of Illinois, Urbana, IL, 2009

1173 DOI: 10.1021/acs.iecr.5b02918


Ind. Eng. Chem. Res. 2016, 55, 1164−1173

Vous aimerez peut-être aussi