Downloaded from http://paperhub.

ir

Available online at www.sciencedirect.com

Procedia Computer Science 18 (2013) 787 – 796

International Conference on Computational Science, ICCS 2013

Determination of the kinetic constants of a chemical reaction in

heterogeneous phase using parameterized metaheuristics
José-Matı́as Cutillas-Lozano∗, Domingo Giménez
Departamento de Informática y Sistemas, Universidad de Murcia, 30071 Murcia, Spain

Abstract
Kinetic parameters of a chemical reaction can be determined using numerical metods by integrating the equation of the reaction
rate. For example, the reaction in the human stomach when neutralizing the acid with an antacid tablet can be simulated by
integrating the equation for the dissolution rate of calcium carbonate present in the tablet together with the joint resolution of
the balances for determining at any time the concentration of all species present in the solution. So, there is a classical opti-
mization problem to determine the values of the kinetic parameters that best fit with the experimental data for a given reaction
mechanism. The optimization problem can be afforded with metaheuristic methods. To obtain a satisfactory metaheuristic it is
necessary to develop and experiment with various methods and to tune each for the problem. The use of a unified parameter-
ized scheme for metaheuristics facilitates the development of metaheuristics and their application. In this paper the application
of parameterized metaheuristics to the problem of determination of kinetic parameters is analysed. Local and global search
methods (GRASP, Tabu Search, Genetic algorithm and Scatter Search) and combinations of them are applied for the simulation
of the above-mentioned reaction. The use of a parameterized metaheuristic scheme allows us to easily experiment and tune
different metaheuristics, so improving the results obtained with a basic optimization method.
Keywords: Chemical reaction, kinetic constants, parameterized metaheuristic schemes

1. Introduction
This paper focuses on estimation with metaheuristic methods of the kinetic parameters of a chemical reaction.
The simulation of the processes that happens in the human stomach when neutralizing the acid with an antacid
tablet is used as proof of concept. A chemical reaction in heterogeneous phase that takes place in a batch stirred
tank reactor is studied [1, 2], and the dissolution process of an antacid tablet is simulated. It is a reaction combined
with mass transfer of carbonate ions present in the solid phase upon contact with an acid solution. The solid
phase consists of an antacid tablet, which contains a certain amount of calcium carbonate. The reaction medium
represents the approximate pH of the human stomach (with values between 3 and 4). The kinetic parameters
of the reaction (order and combined reaction rate constant) are determined using the Euler numerical method
for integrating the equation for the dissolution rate of calcium carbonate in the tablet. Together with the joint
resolution of the balances, the concentration of all species present in the solution are determined at any time. The
effect of different parameters like stirring speed, the initial pH of the medium, the total amount of acid and base
and the state of aggregation of the tablet are studied.

E-mail address: {josematias.cutillas,domingo}@um.es.

∗ Correspondingauthor.

1877-0509 © 2013 The Authors. Published by Elsevier B.V.

Selection and peer review under responsibility of the organizers of the 2013 International Conference on Computational Science
doi:10.1016/j.procs.2013.05.243
 Downloaded from http://paperhub.ir

788 José-Matías Cutillas-Lozano and Domingo Giménez / Procedia Computer Science 18 (2013) 787 – 796

We have a classical optimization problem consisting of the estimation of the kinetic parameters that best fit
with the experimental data for a given reaction mechanism [3, 4]. The solution of the problem can be approximated
with available optimization tools, for example the “Solver” in Excel. The results obtained are used as starting
point, and they make it possible to assert on the order of magnitude of the kinetic constants, allowing us to
establish the search domain for metaheuristic methods, which are used to improve the initial results.
Metaheuristics have been successfully applied to non-linear optimisation problems in many dimensions where
more traditional methods are often found to fail. Also, deterministic, gradient-based optimisation methods do
not search the complete parameter space and can tend to converge towards local extremes of the fitness function,
which is clearly unsatisfactory for problems where the fitness varies non-monotonously with the parameters. On
the other hand, metaheuristics are able to depart from local optima due to their inherent randomness.
The use of a unified parameterized scheme for metaheuristics facilitates the development of metaheuristics
and their application by reusing the basic functions. The scheme has been applied successfully to different prob-
lems [5]: obtaining satisfactory Simultaneous Equation Models from a set of values of the variables, a tasks-
to-processes assignation problem with independent tasks and memory constrains, and an optimization of power
consumption in operation of wells [6]. The parameterized metaheuristic scheme is used and adapted here to the
problem of estimation of the kinetic parameters in a chemical reaction. Four pure metaheuristics (GRASP, Tabu
Search, Genetic algorithm and Scatter Search) are included in the parameterized scheme, which allows us to ex-
periment with the basic metaheuristics and their combinations/hybridizations. Experiments have been conducted
with a total of fifteen metaheuristics (basics and combinations); the initial solution is improved and the preferred
method for this problem is analysed.
The rest of the paper is organized as follows. Section 2 summarizes the problem to be solved and describes
the terminology used. Section 3 shows the general parameterized scheme of metaheuristics and its application to
the problem in hand. The experimental results are shown in Section 4. Section 5 summarizes the conclusions and
proposes future work lines.

2. A problem of determination of the kinetic constants of a chemical reaction in heterogeneous phase

When chemical reactions occur in heterogeneous phase the variables that affect the reaction rate are not only
temperature, pressure or composition. These systems have more than one phase and the problem is more complex:
it may be that the materials move from one phase to another during the reaction. In this case, the mass transfer
rate becomes important and should be included in the terms of the reaction rate.
When reaction rates in complex systems are compared or combined, we have to take into account that if the
change of property (concentration, temperature, ...) is effected by several parallel paths independent of each other,
the overall rate is simply the sum of all individual rates.

2.1. Dissolution of calcium carbonate

According to [7], the kinetic of dissolution of calcium carbonate is a function of the concentration of the
various carbonate species in the solution and, therefore, a function of the partial pressure of carbon dioxide and
pH. A model for the dissolution of calcium carbonate was developed in [8]. According to this model, depending
on the pH, there are four different ways in which the dissolution of calcium carbonate occurs. At pH values lower
than 3.5 the speed of the reaction is proportional to the concentration of H3 O+ according to reaction path 1:

CaCO3 + H3 O+ ↔ Ca2+ + HCO−3 + H2 O (1)

where the dissociation of acetic acid and carbonic and bicarbonate equilibriums is present due to an acidic envi-
ronment and to their being open to the atmosphere.
At these pH values, a dependence of the reaction rate with the stirring speed is observed. This indicates that
ion transportation is the controlling factor in the speed of the process.
At higher pH values the reaction rate becomes less pH dependent, but more dependent on the partial pressure
of carbon dioxide (reaction path 2). In this case the process is controlled by the transport of ions and by the
reaction itself [9]:
CaCO3 + H2CO3 ↔ Ca2+ + 2 · HCO−3 (2)
 Downloaded from http://paperhub.ir

José-Matías Cutillas-Lozano and Domingo Giménez / Procedia Computer Science 18 (2013) 787 – 796 789

The bicarbonate ion formed in this way can not be neutralized as in the previous case, since the pH is not suffi-
ciently acidic.
When the pH is close to saturation, the dissolution rate of calcium carbonate decreases drastically, and starts
to become important hydrolysis reaction (path 3):

CaCO3 + H2 O ↔ Ca2+ + HCO−3 + OH − (3)

Finally, at still higher pH, the reverse reaction of the calcium carbonate precipitation could be added, but given
the nature, concentration and amount of acid to be employed, the path 4 of the precipitation of carbonate may be
neglected.

2.2. Derivation of equations to obtain the kinetic parameters

The reaction medium is quite similar to the human stomach pH, which varies between 3.0 and 4.0. The pH
values of many foods, particularly fruits (apple, orange, peach, etc.) are in this range. Thus, the diluted acetic acid
with a pH of about 3.0 is an effective, inexpensive and reasonable representative of weak organic acids present in
the human body, and it is used as a reactant in the experiments.
As seen, the tablet can be dissolved in three ways: by reaction with acetic acid (equation 1), by reaction with
carbonic acid (equation 2), and by the hydrolysis reaction (equation 3). The overall process speed of dissolution
of the tablet is the sum of the individual speeds of the three ways. Therefore the variation of moles of calcium
over time is:
1 dNCa2+  n
= −k1 an1 H3 O+ 2 − k2 an3 [H2CO3 ]n4 − k3 (4)
V dt
where k1 , k2 and k3 are the combined reaction rate constants and n1 , n2 , n3 and n4 are the reaction orders. The
integration of the equation for a certain value of time t j is:
   
ΔNCa2+  n
= −k1 an1 H3 O+ 2 − k2 an3 [H2CO3 ]n4 − k3 · Δt j (5)
V tj
tj

which gives the increase of calcium in the solution as a function of the concentrations of protons and carbonic
acid, and where a is the area of the tablet, which is known for each increase of time. Therefore the concentration
of Ca2+ in the solution at instant t j is given by:


tj
(ΔNCa2+ )k
Ca2+
=− (6)
tj
k=t
V
0

2.3. Concentration of the different species present in the solution

In order to estimate the validity of the model, general equation 4 is integrated. This requires knowledge of
the concentration of all the species involved in the process at each time. The concentration of the various species
present in solution can be related by a balance of charge:
     
H3 O+ + 2 · Ca2+ = CH3COO− + HCO−3 (7)

where concentrations of hydroxyl and carbonate ions have been neglected because the reaction always proceeds in
an acidic medium. The concentration of Ca2+ (mol/L) shown in these balances can be calculated from the number
of moles of dissolved calcium, calculated by integrating equation 4. The concentration of the other species can be
calculated as a function of the concentration of hydronium ions and the concentration of calcium in the solution.
To calculate the concentration of acetic acid and acetate ion, the material balance can be written as:
 
[CH3COOH]0 = [CH3COOH] + CH3COO− (8)

where [CH3COOH]0 is the initial known concentration of acetic. We also have the equation of the equilibrium
constant of the acetic acid:   
CH3COO− H3 O+
Ka = (9)
[CH3COOH]
 Downloaded from http://paperhub.ir

790 José-Matías Cutillas-Lozano and Domingo Giménez / Procedia Computer Science 18 (2013) 787 – 796

  
and the concentrations of acetic and acetate are calculated as a function of H3 O+ and Ca2+ .
Moreover, as mentioned previously, the amount of carbonate species concentration is at all times in equilibrium
with atmospheric carbon dioxide:

KH H2CO∗3
∗
CO2 + H2 O ←→ H2CO3 KH = (10)
pCO2
This allows us to calculate the concentration of the H2CO3 and HCO−3 species, since [H2CO∗3 ] represents the
concentration of all carbonate species in the solution:
 
H2CO∗3 = [H2CO3 ] + HCO−3 (11)

and these species are in equilibrium as follows:

  

Ka 
HCO−3 H3 O+
H2CO3 + H2 O ←
→ HCO−3 +
+ H3 O Ka = (12)
[H2CO3 ]
These equations allow us to express the concentration of the H2CO3 and HCO−3 species as a function of the
concentration of hydronium ions.

2.4. Determination of kinetic parameters and calculation of the concentration of the species in the solution
For the determination of kinetic parameters and calculation of the concentration of various species with a
rigorous method, a general scheme of calculation must be followed:
1. From the data of the tablet, the dissolution employed and from the acidic constants the initial area, a0 , of
 
the tablet is estimated. The number of moles of calcium carbonate in the tablet initially NCaCO 3 , the initial

acetic acid concentration ([CH3COOH]0 ) and carbonic acid concentration in the medium H2CO∗3 are
 
also estimated. Obviously, at t = 0 the concentration of calcium in the solution Ca2+ is zero.
2. A time increment is fixed and some starting values for the kinetic constants ki and ni appearing in equation
4 are assumed.
 
3. The hydronium ion concentration
 H3 O+ is calculated by solving equations 6-12.
 
4. From H3 O+ cal and H2CO∗3 , the concentration of carbonate species are calculated from equations 11 and
12.
 
5. Known [H2CO3 ] and H3 O+ cal , the value of the initial area of the tablet (a0 ) and with the values of the
kinetic constants supposed, equation 8 is integrated analytically, for the first time increment. This allows us
to calculate the number of moles of calcium that are passing into the dissolution and, hence, the Ca2+ at
that instant
 (equations
5 and 6).
   
6. With Ca2+ and H3 O+ cal , the concentrations of the other species ([CH3COOH] and CH3COO− ) are
calculated.
7. We proceed in the same way for all the time steps. The calcium concentration calculated in the previous
point is used to calculate the area and the concentration of the different species in the following time step.
8. To achieve the parameter combination that best fits the kinetic model in an experiment we must minimize
the following objective function, subject to some restrictions explained in the following subsection:

N  2
pHexp,i − pHcal,i , x ∈ R+
7
minimize O.F. = f (x) = (13)
i=1
 
where x is a real vector containing the kinetic constants and pHexp,i and pHcal,i (calculated from H3 O+ cal,i )
represent the experimental and calculated pH at each moment of the simulation. The difference should be
minimized for the sum of N points of the experiment.
The effect of different parameters, like stirring speed, the initial pH of the medium, the total amount of acid and
base and the state of aggregation of the tablet are studied. This gives different sets of experiments, each with its
theoretical model.
 Downloaded from http://paperhub.ir

José-Matías Cutillas-Lozano and Domingo Giménez / Procedia Computer Science 18 (2013) 787 – 796 791

2.5. Elements and restrictions in the metaheuristic

Using the notation for evolutionary algorithms, an individual is represented by a real vector of size seven,
which is a set of the kinetic constants of the model. A set of individuals constitutes a population. As mentioned in
previous subsections, the kinetic parameters are the combined reaction rate constants k1 , k2 and k3 and the reaction
orders n1 , n2 , n3 and n4 . The ranges of values for these constants are set following empirical criteria. The initial
results obtained with “Solver” for a first range of values on which to experiment are taken into account. These
initial ranges were then extended to obtain upper and lower bounds for both ki and ni . Outside these limits no
values were obtained for the constants in any of the subsequent test runs of the algorithm. So the range for ki is
10−10 , 10 and for ni is [0, 5].
Every time we have to evaluate the fitness of an individual, we must solve the whole chemical system described
in the previous subsection. So the fitness function coincides with the objective function to minimize (equation 13).
The calculation of the fitness function is presented in algorithm 1, with linear complexity O(N), where we have
a loop to N time intervals that calculates the value of each variable of the chemical problem for next time instant
i + 1 depending on the value of the variable in the previous instant i. We can also calculate in this loop the fitness
that matches the O.F. described previously.

Algorithm 1 Calculation of the fitness function for an individual

Calculate Fitness(k1 , k2 , k3 , n1 , n2 , n3 , n4 ):
for i = 0 → N do  
Calculate at instant i: Ca2+ , a, [H3 O+ ] , [HCO− ] , [H2 CO3 ] , pHcal , Δ Ca2+ , [CH3 COOH] , [CH3 COO− ]
 2
Fitness = Fitness + pHexp,i − pHcal,i
end for

When performing the simulations, there are constraints that should be met by both the kinetic constants and
the problem variables. Firstly, the main reaction path should be respected in each pH range. This is achieved by
imposing the constraint that the first term of equation 5 is greater than the second at lower pH, and the first and
second terms are always greater than the third. We thus ensure that the solutions explored make physical sense
and respect the main reaction path in each pH range. Secondly, all the concentrations of chemical species must be
positive in each experimental point. If any negative concentration for some combination of kinetic parameters is
obtained for any chemical species at some time step, the individual becomes unfeasible, and a different individual
is generated.

3. A unified parameterized scheme of metaheuristics

For the application of different metaheuristics to an optimization problem, a good option is to use a common
scheme for several techniques [10]. This approach has been used for other problems [5], and here we study its
adaptation and application to the estimation of kinetic parameters. The use of a common scheme allows to reuse
the functions of some techniques for others. Moreover, to adapt a metaheuristic to a particular problem several
implementations of the functions of the scheme are considered, and these implementations may be used in the
adaptation of different metaheuristics. For the adaptation, parameters in the unified metaheuristic scheme are used
to obtain different combinations of a set of basic metaheuristics and they are evaluated to obtain the parameter
combinations (and consequently the hybridation/combination of metaheuristics) that provide satisfactory solu-
tions. Some of these parameters are common to several metaheuristics. For example, the number of elements in
the initial set, the number of iterations without improving the best solution in the end condition and so on. Hence,
we study the application of different metaheuristics to the optimization problem of determination of the kinetic
constants of a chemical reaction. In order to simplify the experimentation, functions are reused in the unified
parameterized scheme in the algorithm 2, where S , S S , S S 1 and S S 2 represent the sets generated at each stage
of the algorithm and ParamX are the metaheuristic parameters specific to each function.

3.1. Identification of metaheuristic parameters

The basic metaheuristics used to conform the parameterized scheme are Genetic algorithm (GA), Scatter
Search (SS), GRASP (GR) and Tabu Search (TS). From these the set of metaheuristic parameters is established:
 Downloaded from http://paperhub.ir

792 José-Matías Cutillas-Lozano and Domingo Giménez / Procedia Computer Science 18 (2013) 787 – 796

Algorithm 2 Unified parameterized scheme of metaheuristics

Initialize(S ,ParamIni)
while (not EndCondition(S )) do
S S =Select(S ,ParamS el)
S S 1=Combine(S S ,ParamCom)
S S 2=Improve(S S 1,ParamImp)
S =Include(S S 2,ParamInc)
end while

• Initialize: Valid random elements are created to form the initial set with INEIni elements. A smaller subset
with FNEIni elements is usually selected. In some metaheuristics some of the initial elements are improved.
The parameter PEIIni indicates the percentage of elements to improve. The improvement may be more or
less intense, which is represented by an intensification parameter, IIEIni. The parameter S T MIni is used
for the extension of Tabu short-term memory in the initialization improvement.
• EndCondition: The end condition consists of a maximum number of iterations (MNIEnd) or a maximum
number of iterations without improving the best solution (NIREnd).
• Select: The elements can be grouped into two sets, the best and worst according to the objective function.
The number of best elements will be NBES el and of worst elements NWES el, and normally NBES el +
NWES el = FNEIni.
• Combine: The total number of elements to obtain by combination is 2·(NBBCom+NBWCom+NWWCom),
where the three parameters represent the number of combinations among the best elements, among the best
and worst elements, and among the worst elements.
• Improve: As in the improvement in the initialization, PEIImp, IIEImp and S MIImp represent the percent-
age of elements, the intensification of improvement and the short-term memory in the improvement of the
elements generated in the combination, and PEMImp, I MEImp and S MMImp represent the corresponding
values in the mutation.
• Include: The NBEInc best elements are maintained in the reference set, and the other FNEIni − NBEInc
are selected from the remaining elements. LT MInc is a Tabu parameter (long-term memory) that allows the
tracking of the most frequently explored individuals.

Thus, we have a set of 20 parameters with which it is possible to experiment, hybridize, mix and adapt the
metaheuristics to the target problem.
The above functions are general and serve for any optimization problem. If a particularized metaheuristic
for our problem of determination of kinetic constants is wanted, some special considerations should be taken
into account. Due to the large range of real values that can take the constants ki , it is opportune to note that the
mutation of the individual should have special importance in the scheme. So, a good initial variety of individuals,
followed by a maximum percentage mutation of around 20% ensures that a large area of the solution space is
explored. This percentage was set as appropriate after making a series of tests with multiple values using the
parameterized metaheuristic scheme. The initial population was generated with values uniformly spaced along
ranges for ki and ni . For example, if we have ki = a · 10b , a is generated randomly in the real interval [0,10) and b
is generated randomly over the integer interval [-10,0]. The values of ni are generated randomly and uniformly in
the real interval [0,5]. The mutation was to randomly generate a new value for the number of constants selected
by intensity I MEImp.
In the improvement, it is necessary to explore a neighborhood of elements of the order of magnitude of the
considered constant ki . Specifically, for a given value of ki = a · 10b we get a neighbor randomly generating a
new value of a and multiplying or dividing by ten the new ki obtained. For ni , the neighborhood is much more
limited because of the range of values considered. Specifically, the neighbor of ni can be obtained by adding or
subtracting a random percentage of the original value of this constant.
Combinations between two individuals are produced by dividing the individual in two parts. The cutoff is
such that it separates ki from ni . So a descendant can have values of ki from one ascendant and values of ni from
another, or vice versa.
 Downloaded from http://paperhub.ir

José-Matías Cutillas-Lozano and Domingo Giménez / Procedia Computer Science 18 (2013) 787 – 796 793

Because the number of elements that constitutes the individual is not very large (seven kinetic constants), we
should keep in mind that the memory restriction of explored individuals (tabu), should not exceed at most half of
the size of the individual. A value of three might be appropriate for the parameter of tabu memory.

4. Computational results

The simulation of the processes in the dissolution of an antacid tablet has been completed. The evolution of
concentrations of the chemical species studied have been obtained over time. The optimal configuration of kinetic
parameters for the reactions has been sought using metaheuristic methods. Executions have been carried out using
the parameterized scheme of metaheuristics adapting the scheme to our problem. The combinations of kinetic
parameters used in the study are randomly generated. The values used for each parameter for a given combination
of metaheuristics are shown in Table 1, where completion conditions have the typical values N MIEnd = 1000
and NIREnd = 10 and are omitted. Reasonable values of the parameters were chosen after a series of tests in
which they were varied. The combinations shown are those for which the fitness obtained is acceptable without
assuming excessive computational cost.

Table 1. Parameter values for different combinations of metaheuristics considered in the experiments.
GR TS SS GA GR+TS GR+SS GR+GA
Ini INEIni 50 50 20 20 50 50 50
FNEIni 1 1 10 20 1 10 20
PEIIni 100 100 100 0 100 100 100
IIEIni 15 15 15 0 15 15 15
STMIni 0 2 0 0 2 0 0
Sel NBESel 0 1 5 20 0 5 20
NWESel 0 0 5 0 0 5 0
Com NBBCom 0 0 15 10 0 15 10
NBWCom 0 0 20 0 0 20 0
NWWCom 0 0 15 0 0 15 0
Imp PEIImp 0 100 100 0 0 100 0
IIEImp 0 5 5 0 0 5 0
SMIImp 0 2 0 0 0 0 0
PEMImp 0 0 0 20 0 0 20
IMEImp 0 0 0 5 0 0 5
SMMImp 0 0 0 0 0 0 0
Inc NBEInc 0 1 5 20 0 5 20
LTMInc 0 3 0 0 0 0 0

TS+SS TS+GA SS+GA GR+TS+SS GR+TS+GA GR+SS+GA TS+SS+GA GR+TS+SS+GA

INEIni 20 20 20 50 50 50 20 50
FNEIni 10 20 15 10 20 15 15 15
PEIIni 100 0 100 100 100 100 100 100
IIEIni 15 0 15 15 15 15 15 15
STMIni 2 0 0 2 2 0 2 2
NBESel 5 20 8 5 20 8 8 8
NWESel 5 0 7 5 0 7 7 7
NBBCom 15 10 15 15 10 15 15 15
NBWCom 20 0 20 20 0 20 20 20
NWWCom 15 0 15 15 0 15 15 15
PEIImp 100 0 100 100 0 100 100 100
IIEImp 5 0 5 5 0 5 5 5
SMIImp 2 0 0 2 0 0 2 2
PEMImp 0 20 20 0 20 20 20 20
IMEImp 0 2 5 0 5 5 5 5
SMMImp 0 10 0 0 2 0 2 2
NBEInc 5 20 8 5 20 8 8 8
LTMInc 3 3 0 3 3 0 3 3
 Downloaded from http://paperhub.ir

794 José-Matías Cutillas-Lozano and Domingo Giménez / Procedia Computer Science 18 (2013) 787 – 796

Table 2. Values of the problem variables considered in the experiments.

Series S1 S2 S3 S4 S5 S6
N 70 50 34 77 81 69
MT 1.33 1.33 1.31 1.32 1.32 1.30
[HAc]0 · 102 3.98 3.31 2.39 2.18 3.01 1.99
V 250 250 250 100 150 200

GRASP, GA, SS and TS refer to parameters that lead to the metaheuristics GRASP, Genetic Algorithms, Scat-
ter Search and Tabu Search. For example, GR+GA+TS and GR+SS+TS are GRASP methods followed by genetic
algorithms or scatter search with inclusion of Tabu memory, so they are hybrid techniques. GR+GA+SS+TS rep-
resents a GRASP followed by a genetic-scatter search combination with a Tabu memory structure interleaved.
Metaheuristics were applied to six series of experimental data with the configuration of chemical variables
shown in Table 2, where N is the number of experimental points of time and pH to be integrated, MT (g) is the
total mass of the tablet that is dissolving in acid, [HAc]0 (mol/L) is the initial concentration of acetic acid in each
series, and V (mL) is the volume of the solution. Other variables common to all series are as follows: MA  (g) =
0.68 is the active mass of calcium carbonate in the tablet, a0 (cm2 ) = 7.11 is the initial area of the tablet, H2CO∗3
(mol/L) = 10−5 is the total concentration of carbonate species in the solution. And finally, the equilibria constants

involved have values: Ka = 10−4.76 , Ka = 10−6.35 , KH = 10−1.5 .

8 5.2
GR GR
TS TS
SS SS
7 GA 5 GA
GR+TS GR+TS
GR+SS GR+SS
GR+GA GR+GA
6 TS+SS TS+SS
4.8
TS+GA TS+GA
SS+GA SS+GA
5 GR+TS+SS GR+TS+SS
GR+TS+GA 4.6 GR+TS+GA
Fitness

GR+SS+GA GR+SS+GA
Fitness

TS+SS+GA TS+SS+GA
4 GR+TS+SS+GA GR+TS+SS+GA
4.4
3
4.2
2
4
1

3.8
0 0 10 20 30 40 50 60
S1 S2 S3 S4 S5 S6 average Iterations

Fig. 1. (a) Value of the fitness function obtained for different combinations of metaheuristics for different problem series. Results ordered
according to legend order; (b) Optimum values obtained by different metaheuristics in successive iterations for problem S5.

In most cases the final value of the objective function (equation 13) is very similar (figure 1(a)). The lower this
value, the better the result. In general, combinations of metaheuristics produce good fitness values, highlighting
GA metaheuristic and its combinations as the most suitable for our optimization problem. SS and its hybridiza-
tions produce results that are intermediate in most cases. This may be due to the wide range of search managed
for the kinetic constants (specially ki ), that renders local search insufficient to satisfactorily cover the space of
solutions. With GRASP and TS methods the results are worse in most cases, but their hybridization with other
techniques produces satisfactory results. Of course, the preferred method or combination depends on the particular
implementation of the basic functions we are using, but we are not interested in obtaining the best metaheuristic
but in the application of a methodology to easily develop and experiment with families of metaheuristics.
In the last set of data (figure 1(a)) we can see an average of the results for each metaheuristic. It is important to
note the more pronounced deviation of the series S4 and S5 from the average, indicating that the fit of the model
to the experimental data is less accurate in such cases.
Figure 1(b) shows the value of the objective function in successive iterations for each method, for the problem
described in S5 series. We see that metaheuristics including GA have lower fitness values and generally more
iterations to the optimum. On the other hand, we have GRASP, TS, SS and their combinations, with higher fitness
 Downloaded from http://paperhub.ir

José-Matías Cutillas-Lozano and Domingo Giménez / Procedia Computer Science 18 (2013) 787 – 796 795

6 0.0008
pHexp [H3O+]exp
pHcal 0.0007 [H3O+]cal
5.5

0.0006

Concentration (mol/L)
5
0.0005
pH

4.5 0.0004

0.0003
4

0.0002

3.5
0.0001

3 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
Time (min) Time (min)

Fig. 2. (a) Evolution of pH over time. Experimental and calculated values; (b) Experimental and calculated values of concentration of protons
over time.

and, finally, GR+TS+GA with intermediate values of fitness. In GRASP we only have one iteration, represented
with one point on the graph.
It should be noted that the results presented in figure 1(b) are similar to those obtained for other problem sizes,
but only one size is chosen as representative of the rest.
Finally, in figures 2 and 3 we can see the evolution of the experimental and calculated pH and concentrations
of different chemical species given by the model with the optimum kinetic constants obtained for the S5 series.
The final values are: k1 = 6.41·10−5 , k2 = 9.68·10−2 , k3 = 8.87·10−7 , n1 = 0.749, n2 = 1.59·10−2 , n3 = 3.25 and n4
= 1.10. The pH calculated in the simulation is not far from the experimental one especially over ten minutes, and
the prediction made by the kinetic model is acceptable. The concentration of HCO−3 remains at all times below
H2CO3 , which was expected in the pH range considered for series S5 (always less than 6). We can see how the
calcium concentration increases as the tablet is dissolving. Meanwhile, the acetic concentration decreases at the
same rate as the concentration of acetate ion increases, which is expected if the model is correct. In view of these
results, we can get a general idea of the effect that an antacid tablet can have on the human stomach. We see that
stomach acid represented by protons and acetic acid concentrations is gradually reduced. The simulation times
give us a rough idea of the speed with which carbonate acts in the reaction medium, helping to predict the effects
against heartburn. This can be used to design more efficient pills and with relatively low experimental costs.

0.03
1.4e-05 [H2CO3] [Ca2+]
-
[HCO3 ] [CH3COOH]
0.025
1.2e-05 [CH3COO-]
Concentration (mol/L)

Concentration (mol/L)

1e-05 0.02

8e-06
0.015

6e-06
0.01
4e-06

0.005
2e-06

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
Time (min) Time (min)

Fig. 3. (a) Evolution of the concentration of carbonate species over time. Calculated values in the simulation of the reaction; (b) Evolution of
the concentration of acetic species and calcium. Calculated values.
 Downloaded from http://paperhub.ir

796 José-Matías Cutillas-Lozano and Domingo Giménez / Procedia Computer Science 18 (2013) 787 – 796

5. Conclusions and research lines

The kinetic constants of a chemical reaction in heterogeneous phase have been determined satisfactorily using
a unified parameterized scheme for metaheuristics. Initial solutions obtained with the tool “Solver” in Excel have
been improved. The most significant results obtained are:
• The dissolution process of an antacid tablet has been satisfactorily simulated. The evolution of concentra-
tions of the chemical species studied have been obtained over time. The optimal configuration of kinetic
parameters for the reactions has been reached using metaheuristic methods.
• The best results in terms of fitness function are obtained by combining metaheuristics: specifically the
quaternary combination GR+GA+SS+TS and, in general, metaheuristics including GA.
• The effect of SS is less important for this problem because of the wide range of search managed for the
kinetic constants.
As the main conclusion, the use of parameterized schemes of metaheuristics is an appropriate methodology for
the easy design and adaptation of metaheuristics to optimization problems, in particular for those in the study
of kinetic reactions. Statistical tests for comparing different metaheuristics have been avoided because we focus
on the presentation of the methodology for solving the optimization problem. A detailed description of how a
statistical study can be conducted and how to apply it can be found in [5].
Some future research lines we are working or plan to work on are:

• Application of hyperheuristics to obtain the best combination of parameters for the kinetic problem so as to
facilitate the collection of the best metaheuristic or combination. The hyperheuristics can be develop on top
of the same parameterized scheme for metaheuristics [11].
• Application of the same methodology to other optimization problems of chemical parameters and processes.
• For problems with a high computational cost it is convenient to develop unified parameterized schemes in
parallel systems: shared memory (in study [6]), message passing and GPUs.

Acknowledgements

Partially supported by Fundación Séneca, Consejerı́a de Educación de la Región de Murcia, 08763/PI/08, and
the Spanish Ministry of Science, TIN2012-38341-C04-03.

References
[1] O. Levenspiel, Chemical Reaction Engineering, 3rd Edition, Wiley, New York, 1998.
[2] H.S. Fogler, Elements of Chemical Engineering Reaction Engineering, 4th Edition, Prentice Hall, New Jersey, 1999.
[3] S.S. Dhumal, R.K. Saha, Application of Genetic Algorithm for Evaluation of Kinetic Parameters of Coal Pyrolysis, Journal of Energy &
Environment 5, 2006, p. 112.
[4] L. Elliott, D.B. Ingham, A.G. Kyne, N.S. Mera, M. Pourkashanian, C.W. Wilson, Genetic algorithms for optimisation of chemical
kinetics reaction mechanisms, Progress in Energy and Combustion Science 30 (3), 2004, pp. 297–328.
[5] F. Almeida, D. Giménez, J.J. López-Espı́n, M. Pérez-Pérez, Parameterised schemes of metaheuristics: basic ideas and applications with
Genetic algorithms, Scatter Search and GRASP, admitted in IEEE Trans. on Systems, Man, and Cybernectics, Part A: Systems and
Humans.
[6] L.G. Cutillas-Lozano, J.M. Cutillas-Lozano, D. Giménez, Modeling Shared-Memory Metaheuristic Schemes for Electricity Consump-
tion, DCAI, 2012, pp. 33–40.
[7] C.A.J. Appelo, D. Postma, Geochemistry, Groundwater and Pollution, A.A. Balkema, Rotterdam, 1993.
[8] L.N. Plummer, T.M.L. Wigley, D.L. Parkhurst, The kinetics of calcite dissolution in CO2 –water systems at 5 to 60 o C and 0.0 to 1.0 atm
CO2 , American Journal of Science 278, 1978, pp. 179–216.
[9] D. Rickard, E.L. Sjberg, Mixed kinetic control of calcite dissolution rates, American Journal of Science 283, 1983, pp. 815–830.
[10] G.R. Raidl, A unified view on hybrid metaheuristics, Hybrid Metaheuristics Third International Workshop, LNCS 4030, 2006, pp. 1–12.
[11] J.M. Cutillas-Lozano, D. Giménez, L.G. Cutillas-Lozano, Modelling parameterized shared-memory hyperheuristics for auto-tuning, 12th
International Conference Computational and Mathematical Methods in Science and Engineering, 2012, pp 389–400.

Powered by TCPDF (www.tcpdf.org)