Environmental Engineering and Management Journal February 2010, Vol.9, No.

2, 239-247
http://omicron.ch.tuiasi.ro/EEMJ/

“Gheorghe Asachi” Technical University of Iasi, Romania

______________________________________________________________________________________________

OPTIMIZATION BY NN – GA TECHNIQUE OF THE METAL

COMPLEXING PROCESS.
POTENTIAL APPLICATION IN WASTEWATER TREATMENT

Ciprian-George Piuleac1, Silvia Curteanu1, Maria Cazacu2

1
“Gheorghe Asachi” Technical University Iasi, Department of Chemical Engineering,
71 Mangeron Blvd., 700050, Iasi, Romania
2
“Petru Poni” Institute of Macromolecular Chemistry, Iasi, Romania

Abstract

Some samples consisting in silica functionalized in different degrees with dihydroxy-azomethine groups have been tested as
sorbents for copper (II) ions from aqueous solutions in different conditions. An indirect monitoring procedure for the complexing
process was approached: UV-VIS absorption intensity – time variation for the metal-containing solution in which complexing
silica was soaked. In order to find the conditions in which the maximum efficiency in the metal absorption can be obtained
(maximum decreasing of the UV absorbance), a method based on neural networks and genetic algorithms (NN-GA) was applied.
The model included into the optimization procedure was a feedforward neural network with one hidden layer. The optimization
problem was solved with a standard genetic algorithm, which computed the optimal values for reaction conditions (CuCl 2 solution
concentration, pH, complexing groups density in silica) leading to minimum relative absorbance of the metal in shortest time.

Key words: complexed metals, functionalized silica, genetic algorithm optimization, neural network modeling

1. Introduction modifiers like amino or chloro- and silica support or

Chelating polymers play an important role in
the decontamination of aqueous systems such as
wastewater treatment, sludge and noble metal
recovery (Liu and Cheng, 1993). Chelating groups
immobilized on the surface of inorganic solid
supports are gained increasing importance as low cost
sorbents for separation and preconcentration of metal
ions. Advantages of such supports are well
documented (Singh and Mishra, 1991).
Silica gel, the inorganic polymer containing
internal siloxane groups (Si-O-Si) and silanol groups
(Si-OH) distributed on the surface, is currently being
used as a support for various functional groups (Dey
et al., 2006). When these possess complexing ability,
then such silica gel can be used to retain metal traces
from wastewater (Bowe and Martin, 2004; Dudler et
al., 1987). The immobilization of chelating moieties
containing donor atoms on the silica gel surface can
occur via chemical bond formation between organic
through simple physical adsorption processes
(Mahmoud and Gohar, 2000). The chemical
modification of silica gel surfaces with donor atoms
such as N, S, O and P is primarily aimed to improve
the adsorption and exchange properties of the silica
gel, as well as selectivity towards certain metal ions
(Seshadri and Haupt, 1988). Several new chemically
modified silica gels were synthesized, and applied as
normal or selective solid phase extractor for many
toxic and other metal ions. They were found to be
superior to the conventional ion exchange resins in
their ability to extract and selectively preconcentrate
the heavy metal ions from solutions containing high
concentrations of alkali and alkaline earth metal ions
(Jal et al., 2004; Kocjan and Prezeszlakowski, 1992;
Prado and Airoldi, 2001). Advantages of the modified
solid phase extractors compared with the classical
liquid extractor’s were reported in many references
(Blain and Treguer, 1995; Mahmoud and Gohar,
2000). A number of complexing and chelating


Author to whom all correspondence should be addressed: e-mail: silvia_curteanu@yahoo.com
 Piuleac et al. /Environmental Engineering and Management Journal 9(2010), 2, 239-247
systems immobilized on silica gel are available.
Between them, Schiff bases, having multidentate
coordination sites, are known to form complexes with
transition metal ions readily (Che and Cheng, 1986;
Ueno and Martell, 1955).
Silica xerogels functionalized in different
degrees with hydroxyazomethyne groups linked
through a propylene moiety to silica network were
tested in this paper for the uptake of copper (II) from
aqueous solutions. An indirect monitoring procedure
for the complexing process was approached: UV-VIS
absorption intensity – time variation for the metal-
containing solution in which complexing silica was
soaked.
For complex chemical processes, neural
networks (NN), which posses the ability to recognize
and reproduce cause–effect relationships, has gained
popularity in various areas of environmental
approaching (Karatzas and Kaltsatos, 2007;
Manolakos et al., 2007; Mjalli et al., 2007). Neural
networks are efficient modelling tool for complex
processes where the chemical and physical laws are
not completely known because they are black boxes
based on input – output data.
Genetic algorthms (GA) are part of the so-
called evolutionary algorithms and represent an
optimization instrument with increasing application in
scientific problems. These kinds of approaching do
not require any piece of information about the search
space area, just being necessary an objective/fitness
function that assigns a value to any solution (Deb,
1999).
The basic algorithm, simple GA (SGA), has
the advantages that it does not require good initial
guesses for the values of the decision variables, like
traditional methods. It uses a population of several
points simultaneously along with probabilistic
operators (reproduction, crossover and mutation),
inspired by natural genetics. Additionally, SGA has
the advantage of using only the values of the
objective functions and not any derivatives, as
required by gradient search techniques (Guria et al.,
2005).
The applicability of these algorithms has
covered a wide variety of problems because of their
flexibility, ease of operation, minimal requirements
and global perspective that they offer (Deb, 2001).
GAs have found various applications in chemical
engineering including process control, gas pipeline
design, pattern recognition of multivariate chemical
data, optimization of reaction rate parameters,
multipurpose chemical batch plant design, and
scheduling (Yan et al., 2003). Details about different
types of genetic algorithms and a series of their
applications in optimization of the chemical processes
have been given in the reviews of Bhaskar et al.
(2000) and Nandasana et al. (2003).
Our group applied GA optimization method
for different chemical systems (Caliman et al., 2008;
Curteanu et al., 2006; Curteanu and Leon, 2007;
Curteanu and Cazacu, 2007).
240
In a previous paper (Piuleac et al., 2008)
details of preparing silica xerogels with different
contents in chelating groups and its use to extract
metal ions (Cu2+) from aqueous solution were
presented. In order to find the conditions in which a
maximum efficiency in the metal absorption can be
reached and to study the evolution in time of the
absorbance as a function of others process parameters
(saturation time, concentration of the solution,
chelating groups content and pH), a series of
experiments have been performed in different
conditions. The acquired data were used for process
modeling based on artificial intelligence instruments.
It was emphasized the option of using simple neural
networks or stacked neural networks with different
architectures to model the wastewater treatment
process based on silica xerogels functionalized with
hydroxyl-azomethine groups.
The main goal of the present paper is to
develop a general procedure based on neural networks
and genetic algorithms, which could be applied to the
complex optimization problems. Neural networks are
used as efficient modeling tools and genetic
algorithms as solving methods of optimization; a
neural model of the process is included in the
optimization procedure. In this herein case study, the
estimation of the metal absorbance is analyzed as a
function of the process parameters: time, CuCl2
solution concentration, values of complexing groups
density in silica and pH values. There is followed a
maximum value of the absorbance decreasing in the
context of using minimum resources. It was
demonstrated that GA optimization procedure is easy
to apply and provides accurate results.
2. Experimental
2.1. Materials
Some solutions differing by copper dichloride
concentration (5.5 and 11 g /L) and pH (4, 5 and 6)
were prepared by using adequate buffer solutions.
Silica functionalized with
dihydroxyazomethine groups (Fig. 1) in different
degrees (Table 1) was prepared, as already described
(Cazacu et al., 2008), by one-pot procedure, both
silica formation and its functionalization occurring in
the same sol-gel system.
O
O - Si N = HC - OH
O
HO
Fig.1. The silica functionalized with dihydroxyazomethine
groups
 Optimization by NN – GA technique of the metal complexing process - potential application in wastewater treatment
Table 1. The functional groups content in the prepared
silica
Sample Functional groups content in silica (mol/g)
FS1 4.35 x 10-4
FS2 8.70 x 10-4
FS3 13.04 x 10-4
2.2. Metal uptake experiments
UV-VIS spectrum was registered for each
CuCl2 solution. A certain amount of functional silica
gel bringing a certain amount of chelating groups was
introduced in such a solution and UV-VIS spectrum
was registered (by using SPECORD M42
spectrophotometer with quartz cells of 1 cm
thickness) at certain time intervals until the maximum
absorbance remains constant. The corresponding
time, saturation time, was denoted as ts, when either
sorption equilibrium was reached or all functional
groups were occupied by metal complexing (Cazacu
et al., 2008).
2.3. Formulation of the optimization problem
The general solution of an optimization
problem implies a dependency of the following four
elements: an accurate model of the process, a
selected number of control variables, an objective
function and a suitable numerical method for solving
the specified optimization problem.
A good process model is a prerequisite for
application in the optimal control strategy.
The feed-forward, multilayered neural network
is the most used type of network because of the
simplicity of the theory, ease of programming, good
results provided and for its character of universal
function in the sense that if topology of the network is
allowed to vary freely it can take the shape of any
broken curve.
One major problem in the development of
neural network model is the determination of the
network architecture, i.e. the number of hidden layers
and the number of neurons in each hidden layer. The
most used method is the trial-and-error, done by
testing several topologies and comparing the
prediction errors. Lower errors indicate potentially
good architectures, i.e. neural network topologies
with chances to train well and to output good results
(Fernandes and Lona, 2005).
Based on the experimental data, a neural
network modeling methodology was developed to
describe the dependency of the absorbance as a
function of reaction conditions: time (minutes), CuCl2
solution concentration (g∙L-1), complexing groups
density in silica (mol∙g-1) and pH. Consequently, 4
inputs corresponding to the 4 variables mentioned
before are presented as input of the neural model,
along to one output – absorbance. Trial and error
method applied for feedforward neural networks leads
to a multilayer perceptron having one hidden layer
with 6 neurons, MLP(4:6:1). Before using, the neural
model was tested in training and validation phases.
Good agreement between experimental data and
neural network predictions (average relative errors
less than 0.3298 % and correlation over 0.9939)
registered against “unseen” data (validation phase).
Maximum absorbance at every time is noted
At and reported to the absorbance of the initial
solution (A0). For a better comparability of the
results, eliminating the influence of the ratio between
sorbent and solution amounts, the relative absorbance,
Ar=At/A0, can be used.
In this study, the optimization problem
includes the neural model, which is represented as:
NN [Inputs: Time, CuCl2 solution concentration,
complexing groups density in silica, pH; Output:
Absorbance] (1)
The vector of decision variables, u, has the
components:
u = [Time, CuCl2 solution concentration, complexing
groups density in silica, pH] (2)
An admissible control input u* should be
formed in such a way that the performance index, J,
defined by the following equation, to be minimized:
J  Ar (3)
subjected to:
u min  u  u max
0  time  2200 min
(4)
5.5  CuCl 2  11 g L1
44  10 5  groups density in silica  13  10 4 mol  g 1
4  pH  6
The absorbance is a measure of the ions
elimination from wastewater, with values between
1.017 and 3.01. The objective function supposes the
minimization of the relative absorbance value, which
means the maximization of the decrease of the
absorbance.
The constraints (4) are very important to
define the range of variation of parameters and to
disregard possible solutions that could be interesting
in a theoretical approach to the problem.
GA posses several characteristics that make
them attractive to solve optimization problems where
the model is only available in an implicit form
(black-box model). Firstly, due to the fact that it is
based on a direct search method, it is not necessary to
have explicit information of the mathematical model
or its derivatives. Secondly, the search for the
optimal solution is not limited to one point, but it
rather relies on several points simultaneously,
therefore the knowledge of initial feasible points is
241
 Piuleac et al. /Environmental Engineering and Management Journal 9(2010), 2, 239-247
not required and such points do not influence the
final solution (Leboreiro and Acevedo, 2004).
The optimization procedure includes the
neural model MLP(4:6:1) and it is solved with a
simple genetic algorithm. Genetic algorithm
provides, after an iterative calculus, the optimal
values for decision variables (time, CuCl2 solution
concentration, complexing groups density in silica,
pH), which are the inputs for the neural network
model. With these inputs, the neural network
computes the final value of the relative absorbance. If
the two values are identical or there is a very tight
difference between them, it can be conclude that the
task of the optimization, represented by minimum of
the objective function, J, is achieved. Fig. 2
illustrates this optimization procedure.
Fig. 2. The optimization method based on NN and GA
2.4. GA optimization technique
Genetic algorithms are intelligent stochastic
optimization techniques based on the mechanism of
natural selection and genetics. GAs start with an
initial set of solutions, called population. Each
solution in the population is called chromosome (or
individual), which represents a point in the search
space. The chromosomes are evolved through
successive iterations, called generations, by genetic
operators (selection, crossover and mutation) that
mimic the principle of natural evolution. A set of
solutions are analyzed and modified by genetic
operations simultaneously, where selection operator
can select some “good” solutions as seeds, crossover
operator can generate new solutions hopefully
retaining good features from parents, and mutation
operator can enhance diversity and provide a chance
to escape from local optima (Wang et al., 2006).
In a GA, a fitness value is assigned to each
individual according to a problem-specific objective
function. Generation by generation, the new
individuals, called offspring, are created and survive
with chromosome in the current population, called
parents, to form a new population.
In the GA model, They were used real values
encoding for the chromosomes. There are other
242
approaches for genetic algorithm based optimization
which use binary solution representation, as it is the
simplest type of encoding, in which chromosomes are
composed only of 1’s and 0’s. Even the number of
alleles is thus rather small (two), this encoding is very
common, because it is very easy to use. However,
value encoding is more general, because genes are
real numbers. Some experiments (Michalewicz, 1996)
have shown that real value encoding is more time
efficient, with better precision of the solutions.
The initial population is generated randomly.
Offspring is created by genetic operators and it is
stored in a population pool that is a collection of
offspring and their parents.
Selection compares the chromosomes in the
population aiming to choose those which will take
part in the reproduction process. The selection occurs
with a given probability on the base of fitness
functions. The fitness function plays a role of the
environment to distinguish between good and bad
solutions. The individuals with higher fitness must
have more chances to reproduce. For each individual
to be created in the next generation, two parents are
thus selected.
There are different methods for selection
phase; our paper uses rank selection which first ranks
the population and then every chromosome receives
fitness from this ranking. The worst will have fitness
1, second worst 2 etc. and the best will have fitness n
(number of chromosomes in the population).
The recombination (crossover) has as main
purpose the recombination the features of two
randomly selected parents from the mating pool with
the aim of producing better offspring. The variant of
crossover used in this study supposes different points
for all genes, which means that the new individual
will no longer be on the line segment that links its
parents. The offspring will look more like one parent
regarding a feature and less regarding another.
After recombination, offspring undergoes to
mutation. Generally, the mutation refers to the
creation of a new chromosome from one and only
one individual with predefined probability. Mutation
is used to produce small perturbations on
chromosomes to promote diversity of the population.
Our GA includes a variant of mutation named
resetting.
A gene value is reset to a random value in its
search interval. The purpose is to refresh the search
process, in case when the genetic diversity of the
population decreases (so no longer converges to the
solution) or the algorithm has converged into a local
optimum. Each gene is independently considered,
and mutation gives it a new random value in the
initialization interval. Only some genes change
(possibly all, but unlikely).
After the three operators are carried, the
offspring is inserted into the population, replacing
the parent chromosomes in which they were derived
from, producing a new generation. The best
individual is copied directly into the new population
(the elitism technique) and the rest of the individuals
 Optimization by NN – GA technique of the metal complexing process - potential application in wastewater treatment

are replaced by the new generations. So, in order to 1.02

keep the best solution, there were considered an
elitism factor fe = 1, that is the best individual is 1.00
copied directly in the new generation. That ensures G1_exp

Relative absorbance
0.98
that the overall solution of the GA will not get worse. G1_net
The stopping criterion determines when GA 0.96
G2_exp
G2_net
will break. In other words, the genetic operations are
G3_exp
repeated until a termination condition is finding. In 0.94
G3_net
this implementation GA was stopped if the maximum 0.92
G4_exp
number of generations has been executed or the pre- G4_net

set number of generations without improvement in the 0.90

last best solution has been reached.
0.88
0 500 1000 1500 2000 2500
3. Results and discussion
Time, min

Due to the presence in positions favourable to

closing the chelate ring, the azomethine and hydroxyl Fig. 4. Evolution of the relative absorbance in time
groups attached to the silica can co-ordinate the metal (samples G1, G2, G3, G4) obtained experimentally and
predicted with MLP(4:6:1) on training data
ions from an aqueous solution forming complex like
that showed in Fig. 3 and proved by the spectral
methods (Piuleac et al., 2008). The Si-OH groups on 1.02
the surface permit the silica wetability and the metal 1.00
ion access to the chelating groups (Piuleac et al.,
0.98
2008). G5_exp
Relative absorbance

0.96 G5_net
G6_exp
0.94 G6_net
0.92 G7_exp
O
G7_net
0.90 G8_exp
O - Si N = HC - OH
0.88 G8_net
O
O Cu O 0.86
O
0.84
HO - CH = N Si - O 0 200 400 600 800 1000 1200 1400 1600

Time, min
O

Fig. 5. Evolution of the relative absorbance in time

Fig.3. The copper complexed by functionalized silica (samples G5, G6, G7, G8), obtained experimentally and
predicted with MLP(4:6:1) on training data
Three variables were involved in this process:
functional groups content in silica, metal
1.1
concentration and pH of the solutions. UV spectrum
for each initial solution was registered, the maximum
1.0
absorbance being A0. A determined amount of
functional silica gel was introduced in such a solution
Relative absorbance

and UV spectrum was registered at certain time 1.0

Experimental
intervals. Thus, the concentration of the copper ions Modeling
in solution decreases and, by consequence, the UV 0.9
absorbance also decreases Figs. 4 and 5 present some
examples. Maximum absorbance at every time (At) 0.9
was determined from the spectra as an indirect
measure of the metal content and reported to the 0.8
absorbance of the initial solution (A0). The time at 1.0 3.0 5.0 7.0 9.0 11.0 13.0 15.0 17.0 19.0 21.0 23.0 25.0
which the absorbance remained constant was Exemples
considered the saturation time.
The good agreement between experimental and
simulation data obtained with neural model in Fig. 6. Relative absorbance obtained experimentally
and predicted with MLP(4:6:1) on testing data
validation phase is shown in Fig. 6. The neural
network MLP(4:6:1) has a good capacity of
Population size, number of generations,
generalization, therefore it can be used trustily in the
crossover probability and mutation probability are
optimization procedure.
known as the control parameters of genetic

243
 Piuleac et al. /Environmental Engineering and Management Journal 9(2010), 2, 239-247

algorithms. The values of these parameters must be
specified before the execution of GA and they depend
on the nature of the objective function. One important
thing in using GA as solving method is to adjust its
parameters to the particular problem approached, so
as to obtain good solutions and to preserve the
algorithm from a preliminary convergence.
The fitness function in GA is the objective
function of the optimization procedure. The results of
the optimization are represented by the values of the
decision variables (time, CuCl2 solution
concentration, complexing groups density in silica,
pH) that lead to a minimum value of the objective
function, which means the achievement of minimum
value for the relative absorbance (maximum decrease
of the absorbance).
The optimization procedure is implemented in
Matlab 7.5 with original software, as specific
functions were programmed for each phase of the
genetic algorithm.
Since GA is a stochastic algorithm, it was run
for many times, for each situation (each row in the
tables of results) to get statistically meaningful values
for the time taken. Longer runs are not necessary as
GA is already a population based procedure that
works with multiple solutions. One of the solutions
(the best solution) is chosen and inserted into the
results Tables.
Table 2 shows the influence of initial
population. There is known that the grower parameter
is the better results of optimization are obtained, but
this kind of rule is applied until one certain limit.

Table 2. The influence of population dimension on the performance of GA optimization

No. Dimension of Number of GA method Decision variables Values of objective

population generation function
Time = 2125.4214
1 50 100 Cross rate =0.4 Conc = 5.5007
Mutation rate=0.04 Gr_dens = 0.0010 0.9433
pH = 5.5017

Time = 2199.3954
Conc = 5.5111
2 100 100 Gr_dens = 0.0010 0.9427
pH = 5.5008

Time = 2173.9741
Conc = 5.5024 Gr_dens =
3 150 100 0.0010 0.9428
pH = 5.5003

Time = 2194.5754
Conc = 5.5082 Gr_dens =
4 200 100 0.0010 0.9429
pH = 5.5018

Time = 2190.0300
Conc = 5.5062
5 300 100 Gr_dens = 0.0010 0.9428
pH = 5.5005

Time = 2187.1616
Conc = 5.5092 Gr_dens =
6 400 100 0.0010 0.9429
pH = 5.5015

Time = 2190.0520
Conc = 5.5039 Gr_dens =
7 500 100 0.0010 0.9429
pH = 5.5032

Time = 2198.7309
Conc = 5.5191
8 600 100 Gr_dens = 0.0010 0.9429
pH = 5.5024

Time = 2186.6470
Conc = 5.5005 Gr_dens =
9 800 100 0.0010 0.9429
pH = 5.5023

Time = 2193.6061
Conc = 5.5072
10 1000 100 Gr_dens = 0.0010 0.9427
pH = 5.5006

244
 Optimization by NN – GA technique of the metal complexing process - potential application in wastewater treatment

Starting with a dimension of initial population
of 50, the results obtained are good, although at high
values there are obtaining better results. The
improving is becoming insignificant, so the growth of
the execution time doesn’t justify. For this case, the
relative absorbance is around 0.94, the values of time
are at superior limit of the considered interval
(experimental time interval), the CuCl2 concentration
about 5.5, pH = 5.5 and the group density around
1·10-3 (relatively a high value).
In Table 2, Conc refers to CuCl2 solution
concentration (g∙L-1) and Gr_dens represents
complexing groups density in silica (mol∙g-1).
A similar problem is about the influence of the
number of generations (Table 3). Taken into account
a range between 50 and 1000 generations with a
dimension of population established at 50, considered
as an optimal value, the results seem to be similar in
all the cases. The influence of crossover and mutation
rates was also studied, by varying the crossover rate
between 0.2 and 2 and the mutation rate between
0.001 and 1. The best values for these two
parameters were 0.9 and 0.03, respectively. The
decreasing of the cross rate could lead at worse
results. Consequently, the best GA parameters for the
approached case study were: dimension of initial
population = 50, number of generations = 50,
crossover rate = 0.9 and mutation rate = 0.03.
Running GA optimization with these
parameters, the following results were obtained: time
= 2100 min, CuCl2 concentration = 5.85 g·L-1, group
density = 0.0010 mol·g-1 and pH = 5.5, with a
maximum value of relative absorbance of 0.945.
Generally, the optimization results are not
significantly influenced by the values of GA
algorithm. Taking into account the elitism included in
the optimization procedure, this technique does not
allow the solution becomes worse.
On the other hand, there is compensation
between the parameters values; for instance, if the
crossover or mutation rates are unsuitable, high
values for initial population dimension and number of
generations are necessary to reach optimal solution.
It could be better to obtain a maximum
decrease of the absorbance in the shorter time. That
why several runs were performed narrowing the time
interval (Table 4). One can see that the value of
relative absorbance increases as the time interval
becomes narrower. The dependence between the
values of the relative absorbance and the optimal
value of time is given in Fig. 7.

Table 3. The influence of number of generation on the performance of GA optimization

No. Dimension of population Number of generation GA method Decision variables

1 50 50 Time = 2184.2895
Cross Rate=0.8 Conc = 5.5595
Mute Rate=0.04 Gr_dens = 0.0010
pH = 5.5214
2 50 100 Time = 2147.1225
Conc = 5.5066
Gr_dens = 0.0010
pH = 5.5096
3 50 150 Time = 2193.0200
Conc = 5.5207
Gr_dens = 0.0010
pH = 5.5044
4 50 200 Time = 2198.6052
Conc = 5.5064
Gr_dens = 0.0010
pH = 5.5076
5 50 300 Time = 2190.5278
Conc = 5.5027
Gr_dens = 0.0010
pH = 5.5010
6 50 400 Time = 2194.4349
Conc = 5.5051
Gr_dens = 0.0010
pH = 5.5002
7 50 600 Time = 2198.6412
Conc = 5.5003
Gr_dens = 0.0010
pH = 5.5003
8 50 800 Time = 2199.9137
Conc = 5.5005
Gr_dens = 0.0010
pH = 5.5015
9 50 1000 Time = 2196.8627
Conc = 5.5018
Gr_dens = 0.0010
pH = 5.5002

245
 Piuleac et al. /Environmental Engineering and Management Journal 9(2010), 2, 239-247

Table 4. Optimization with different restrictions imposed to the time interval

No Imposed Dimension. of No. of GA Method Decision variables Output of Values of objective

interval of iniţial generations neural model function
time population
1 0÷2200 50 50 Rata_cross =0.9 Time = 2043.6148 0.9438 0.9438
Rata_mut=0.03 Conc = 5.5002
Gr_dens = 0.0010
pH = 5.5005
2 0÷2000 Time = 1997.6259 0.9442 0.9442
Conc = 5.5060
Gr_dens = 0.0010
pH = 5.5007
3 0÷1800 Time = 1798.9703 0.9471 0.9471
Conc = 5.5035
Gr_dens = 0.0010
pH = 5.5001
4 0÷1000 Time = 265.0295 0.9518 0.9518
Conc = 6.9988
Gr_dens = 0.0010
pH = 5.9999
5 0÷500 Time = 265.1145 0.9518 0.9518
Conc = 6.9996
Gr_dens = 0.0010
pH = 5.9991
6 0÷200 Time = 199.5373 0.9536 0.9536
Conc = 6.9991
Gr_dens = 0.0010
pH = 5.9993
7 0÷150 Time = 149.9525 0.9590 0.9590
Conc = 6.9978
Gr_dens = 0.0010
pH = 5.9998
8 0÷100 Time = 99.9194 0.9692 0.9692
Conc = 5.5005
Gr_dens = 0.0010
pH = 6.0000
9 0÷50 Time = 49.9311 0.9845 0.9845
Conc = 5.5045
Gr_dens = 0.0010
pH = 5.9994

One can see that a minimum relative algorithms and neural networks, applied to a
absorbance is associated with maximum time. wastewater treatment by implying silica xerogel
Consequently, a compromise between the limit compounds.
situations presented in Fig. 5 supposes the values of The genetic algorithm solves the optimization
relative absorbance of 0.950 0.955 with time around problem and the neural network constitutes the model
250 minutes. included in the optimization procedure. Simple
architecture of the neural network is proposed for
0.99 process modeling: a simple feed forward neural
0.985 network with one hidden layer. The simple genetic
0.98 algorithm proves to be a good tool for solving the
Relative absorbance

0.975 optimization problem, providing important

0.97 information useful in experimental practice.
0.965 The method can be easily extended and
0.96
adapted to other environmental oriented processes,
0.955
with high chances of providing accurate results by
0.95
simple handling.
0.945
0.94
0 500 1000 1500 2000 2500 Acknoledgements
Time, min This work was done by financial support provided by
Romanian Ministry of Education and Research through
National Program for Research, Development and
Fig. 7. Relative absorbance vs. time at optimum point Innovation II, project 71-006/2007.

4. Conclusions References

This paper provides a general and simple Bhaskar V., Gupta S.K., Ray A.K., (2000), Applications of
optimization methodology, based on genetic multiobjective optimization in chemical engineering,
Reviews in Chemical Engineering, 16, 1-54.

246
 Optimization by NN – GA technique of the metal complexing process - potential application in wastewater treatment
Blain S., Treguer P., (1995), Iron (II) and iron(III)
determination in sea water at the nanomolar level with
selective on-line preconcentration and
spectrophotometric determination, Analytica Chimica
Acta, 308, 425-432.
Bowe C.A., Martin D.F., (2004), Extraction of heavy metals
by 2-mercaptoethoxy groups attached to silica gel,
Journal of environmental science and health. Part A,
Toxic/hazardous Substances and Environmental
Engineering, 39, 1479-148.
Cazacu M., Vlad A., Airinei A., Alexandru M., (2008),
Silica Encapsulating Lanthanum Complexes by Sol-
Gel Technique, Polymers International, in press.
Che C.M., Cheng W.K., (1986), Manganese(III) amide
complexes as a new class of catalyst for efficient
alkene epoxidation, Journal of Chemical Society,
Chemical Communications, 1443-1444.
Curteanu S., Leon F., Galea D., (2006), Alternatives for
Multiobjectives Optimization of a Polymerization
Process, Journal of Applied Polymer Science, 100,
3680-3695.
Curteanu S., Cazacu M., (2007), Neural networks and
genetic algorithms used for modeling and optimization
of the siloxane-siloxane copolymers synthesis, Journal
of Macromolecular Science, Part A Pure and Applied
Chemistry, A45, 23-36.
Curteanu S., Leon F., (2007), Optimization strategy based
on genetic algorithms and neural networks applied to a
polymerization process, International Journal of
Quantum Chemistry, 108, 617-630.
Caliman F., Curteanu S., Betianu C., Gavrilescu M.,
Poulios I., (2008), Neural networks and genetic
algorithms optimization of the photocatalytic
degradation of Alcian Blue 8GX, Journal of Advanced
Oxidation Technologies (ACS), 11, 316-326.
Deb K., (1999), An introduction to Genetic Algorithms,
Sadhana-Academy Proceedings in Engineering
Sciences, 24, 293-315.
Deb K., (2001) Multiobjective Optimization Using
Evolutionary Algorithms, Chichester, Wiley.
Dey R. K., Jha U., Singh A.C., Samal S., Ray A.R., (2006),
Extraction of metal ions using chemically modified
silica gel covalently bonded with 4,4′-
diaminodiphenylether and 4,4`-
diaminodiphenylsulfone-salicylaldehyde Schiff Bases,
Analytical Sciences, 22, 1105-1110.
Dudler V., Lindoy L.F., Sallin D., Schlaepfer C.W., (1987),
An oxygen-nitrogen donor macrocycle immobilized on
silica gel. a reagent showing high selectivity for
Copper(II) in the presence of Cobalt(II), Nickel(II),
Zinc(II) or Cadmium(II), Australian Journal of
Chemistry, 40, 1557 – 1563.
Fernandes, F.A.N., Lona, L.M.F., (2005), Neural network
applications in polymerization processes, Brazilian
Journal of Chemical Engineering, 22, 323-330.
Guria C., Bhattacharza P.K., Gupta S.K., (2005), Multi-
objective optimization of reverse osmosis desalination
units using different adaptations of the non-dominated
sorting genetic algorithm (NSGA), Computers and
Chemical Engineering, 29, 1977-1995.
Jal P.K., Patel S., Mishra B.K., (2004), Chemical
modification of silica surface by immobilization of
functional groups for extractive concentration of metal
ions, Talanta, 62, 1005-1028.
Karatzas K.D, Kaltsatos S., (2007), Air pollution modelling
with the aid of computational intelligence methods in
Thessaloniki, Greece Simulation Modelling Practice
and Theory, 15, 1310-1319.
Kocjan R., Prezeszlakowski S., (1992), Calcon-modified
silica gel sorbent. Application to preconcentration or
elimination of trace metals, Talanta, 39, 63-68.
Leboreiro J., Acevedo J., (2004), Processes synthesis and
design of distillation sequence using modular
simulators: a genetic algorithm framework, Computers
and Chemical Engineering, 28, 1223-1236.
Liu C.-Y., Cheng L., (1993), Chelating Polymers. In:
Facets of Coordination Chemistry, Agarwal B.V.,
Munshi K.N., Dey A.K. (Eds.), World Scientific.
Mahmoud M.E., Osman M.M., Amer M.E., (2000),
Selective pre-concentration and solid phase extraction
of mercury(II) from natural water by silica gel-loaded
dithizone phases, Analytica Chimica Acta, 415, 33-40.
Mahmoud M.E., Gohar G.A., (2000), Silica gel-
immobilized-dithioacetal derivatives as potential solid
phase extractors for mercury(II), Talanta, 51, 77-87.
Manolakos E., Virani H., Novotny V., (2007), Extracting
knowledge on the links between the water body
stressors and biotic integrity, Water Research, 41,
4041-4050.
Michalewicz Z., (1996), Genetic Algorithms + Data
Structures = Evolution Programs, Springer Verlag,
Berlin.
Mjalli F.S., Al-Asheh S., Alfadala H.E., (2007), Use of
artificial neural network black-box modeling for the
prediction of wastewater treatment plants performance
Journal of Environmental Management, 83, 329-338.
Nandasana A.D., Ray A.K., Gupta S.K., (2003),
Applications of the Non-Dominated Sorting Genetic
Algorithm (NSGA) in chemical reaction engineering,
International Journal of Chemical Reactor
Engineering, 1, 1-15.
Piuleac C.G., Curteanu S., Cazacu M., (2008), Stacked
neural network modeling applied to the process of
metal ion sorption by the functional silica xerogel,
Polymer-Plastics Technology and Engineering, in
press.
Prado A.G.S., Airoldi C., (2001), Adsorption,
preconcentration and separation of cations on silica gel
chemically modified with the herbicide 2,4-
dichlorophenoxyacetic acid, Analytica Chimica Acta,
432, 201-211.
Seshadri T., Haupt H.J., (1988), Silica-immobilized 2-[(2-
(triethoxysily)ethyl)thio]aniline as a selective sorbent
for the separation and preconcentration of palladium,
Analytical Chemistry, 60, 47-52.
Singh D.K. ,. Mishra N.K, (1991), Zinc silicate-bonded
diethyldithiocarbamate - a selective adsorbent for
separating transition metal ions and preconcentration
of palladium(II), Chromatographia, 31, 300-302.
Ueno K., Martell A.E., (1955), Infrared study of metal
chelates of bisacetylacetoneethylenediimine and
related compounds, Journal of Physical Chemistry, 59,
998-1004.
Wang L., Zhang L., Zheng D.Z., (2006), An effective
hybrid genetic algorithm for flow shop scheduling with
limited buffers, Computers and Operations Research,
33, 2969-2971.
Yan X.F., Chen D.Z., Hu, S.X., (2003), Chaos-genetic
algorithms for optimizing the operating conditions
based on RBF-PLS model, Computers and Chemical
Engineering, 27, 1393-1404.
247