Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

R7 Ladhari

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/330931180

Parameter identification of a reduced nonlinear model for an activated sludge


process based on cuckoo search algorithm

Article  in  Transactions of the Institute of Measurement and Control · February 2019


DOI: 10.1177/0142331218824384

CITATION READS

1 440

4 authors:

Taoufik Ladhari Intissar Khoja


National Engineering School of Monastir National Engineering School of Monastir
33 PUBLICATIONS   53 CITATIONS    5 PUBLICATIONS   18 CITATIONS   

SEE PROFILE SEE PROFILE

M'Sahli Faouzi Anis Sakly


University of Monastir National Engineering School of Monastir
242 PUBLICATIONS   1,093 CITATIONS    245 PUBLICATIONS   1,542 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

A cascade high gain observer is designed to estimate the first four leading moments of the crystal size distribution (CSD) and the mean crystal size in batch
crystallization processes. View project

Control and analysis of the muscle force response by Functional Electrical Stimulation View project

All content following this page was uploaded by Taoufik Ladhari on 14 February 2019.

The user has requested enhancement of the downloaded file.


Publishing

Page Proof Instructions and Queries


Journal Title: TIM
Article Number: 824384

Thank you for choosing to publish with us. This is your final opportunity to ensure your article will be accurate at publication. Please review
your proof carefully and respond to the queries using the circled tools in the image below, which are available by clicking ‘‘Comment’’ from the
right-side menu in Adobe Reader DC.*

Please use only the tools circled in the image, as edits via other tools/methods can be lost during file conversion. For comments, questions, or
formatting requests, please use . Please do not use comment bubbles/sticky notes .

*If you do not see these tools, please ensure you have opened this file with Adobe Reader DC, available for free at get.adobe.com/reader or by
going to Help >Check for Updates within other versions of Reader. For more detailed instructions, please see us.sagepub.com/
ReaderXProofs.

Sl. No. Query

Please confirm that all author information, including names, affiliations, sequence, and contact
details, is correct.
Please review the entire document for typographical errors, mathematical errors, and any other
necessary corrections; check headings, tables, and figures.
Please ensure that you have obtained and enclosed all necessary permissions for the reproduction
of art works (e.g. illustrations, photographs, charts, maps, other visual material, etc.) not owned
by yourself. Please refer to your publishing agreement for further information.
Please note that this proof represents your final opportunity to review your article prior to publi-
cation, so please do send all of your changes now.
Please confirm that the Funding and Conflict of Interest statements are accurate.
1 change ‘et al.’ to ‘and Gouzé’?
2 change ‘et al.’ to ‘and Gomez-Quintero’?
3 change ‘et al.’ to ‘and Olsson’?
4 change ‘et al.’ to ‘and Georgakakos’?
5 change ‘et al.’ to ‘and Olsson’?
6 not found in the References, but there is one Julien reference missing a date – is this the same
reference?
7 change ‘et al.’ to ‘and Barbu’?
8 insert (PSO) here, as it appears below?
9 change ‘et al.’ to ‘and Chowdhury’?
10 change ‘et al.’ to ‘and Abdelaziz’?
11 change ‘et al.’ to ‘and Zainal’?
12 change ‘et al.’ to ‘and Prabin’?
13 change ‘et al.’ to ‘and Antanassova’?
14 please give full definition of AG
15 see previous query, not found in text
16 change ‘et al.’ to ‘and Gomez-Quintero’?
17 please see previous query, not found in References
18 change ‘et al.’ to ‘and Gomez-Quintero’?
19 change ‘children’ to ‘offspring’?
20 Please note that colour will not appear in the printed version. We suggest using different type
lines so they can be differentiated
21 change ‘et al.’ to ‘and Mead’?
22 should ‘, the’ be deleted here?
23 Please note that colour will not appear in the printed version. We suggest using different type
lines so they can be differentiated
24 Please note that colour will not appear in the printed version. We suggest using different type
lines so they can be differentiated
25 Please note that colour will not appear in the printed version. We suggest using different type
lines so they can be differentiated
26 Please note that colour will not appear in the printed version. We suggest using different type
lines so they can be differentiated
27 Please note that colour will not appear in the printed version. We suggest using different type
lines so they can be differentiated
28 Please note that colour will not appear in the printed version. We suggest using different type
lines so they can be differentiated
29 Please note that colour will not appear in the printed version. We suggest using different type
lines so they can be differentiated
30 Please note that colour will not appear in the printed version. We suggest using different type
lines so they can be differentiated
31 please check full sentence – is something missing?
32 please confirm the publisher’s location and name
33 please confirm the location and date of the conference, and the publisher’s location and name
34 please confirm the volume and issue numbers
35 Is this a conference reference? If so, please confirm the date it took place, the page numbers, and
the publisher’s location and name
36 please confirm the date of the conference, and the publisher’s location and name
37 please confirm the page numbers
38 Please confirm the issue number
39 please confirm the issue number
40 Missing a date. Should this be 1997 – see query regarding Julien, 1997 in the main text
41 please confirm the location and date of the conference, and the publisher’s location and name
42 please confirm the issue number
43 please confirm the volume, issue and page numbers
44 please confirm the page numbers
45 Is this a thesis reference? Please provide more details.
46 please confirm the volume, issue and page numbers
47 please confirm the volume, issue and page numbers
48 please confirm the issue number
49 please confirm the page numbers
50 please confirm the location and date of the conference, and the publisher’s location and name
Article
Transactions of the Institute of
Measurement and Control
Parameter identification of a reduced 1–12
Ó The Author(s) 2019

nonlinear model for an activated Article reuse guidelines:


sagepub.com/journals-permissions
DOI: 10.1177/0142331218824384
sludge process based on cuckoo search journals.sagepub.com/home/tim

algorithm

Taoufik Ladhari, Intissar Khoja, Faouzi Msahli and Anis Sakly

Abstract
Parameter identification plays a key role in systems’ modeling and control. This paper deals with a parameter identification problem for an activated
sludge process used in wastewater treatment. The considered model is a nonlinear one inspired from the well-known ASM1. Nature-inspired algo-
rithms have gained significant attention over the last years as useful means to solve parameter identification problem. The proposed approach in this
paper is the cuckoo search algorithm based on both the fascinating brood parasitic behavior and the lévy flights. The advantages of this method are its
simplicity and robustness, but it requires a good tuning of its parameters to have the best results. The comparison of the simulation results with the
Nelder-Mead method, genetic algorithm, and particle swarm optimization proves the capability of this method to identify the model’s parameters with
high precision.

Keywords
Activated sludge process, parameter identification, cuckoo search algorithm

Introduction All these models are useful and efficient for description,
design and modeling of the activated sludge processes
The majority of mathematical tools investigated in designing, (Gernaey et al., 2004; Hauduc et al., 2010, 2011; Rieger et al.,
solving practical control and identification problems stand in 2001; Wu et al., 2016) but at the same time are complicated
need of simplified models. These models must describe ade- and highly nonlinear for observation and control strategies
quately the system performance and they depend on their uti- that require creating more simpler and intelligible models
lization’s purposes. Thus, a compromise between accuracy (Gomez Quintero et al., 2000; Gonzalez-Miranda et al., 2009;
and simplicity is required. For activated sludge processes, Hadj-Sadok et al.[AQ: 1], 2001; Holenda et al., 2008; Julien
many models have been developed for design, control and et al., 1999; Queinnec et al.[AQ: 2], 2009; Vrečko et al., 2011;
specific analysis aims. Zhang et al., 2008).
All the developed models can be gathered in one class Over the years, many models have been created mainly
called the Activated Sludge Model (ASM) family. This latter is based on the ASM family to accomplish an effective control
proposed by the International Water Association (IWA). It plan while responding to the modifications in the operating
starts with the ASM1 developed in 1982 as the reference model conditions within the treatment system. The distinct reduction
for all the following works. This model presents a general methods are mainly based on: simplification of biomass
acceptance of biological plant modeling. It describes the dynamics, linear approximation of the nonlinear terms, utili-
removal of nitrogen and organic compounds accompanying zation of the available online measurements and so forth
the consumption of oxygen and nitrate (Henze et al., 1987). (Birs et al., 2016; Gomez Quintero et al., 2000; Holenda et al.,
Then, the ASM2 has been created in 1995 by extending the
capabilities of the ASM1 to the biological phosphorus removal
(Gujer et al., 1995). After that, the ASM2d has been built on
its ancestor the ASM2 by adding the denitrification activity in Research Unit: Industrial systems study and renewable energy (ESIER),
order to give a better description for the performance of phos- the National Engineering School of Monastir (ENIM), University of
phate and nitrate (Henze et al., 1999). In 1998, The ASM3 was Monastir, Tunisia
also developed for biological nitrogen removal, with basically
Corresponding author:
the same objective as the ASM1 (Gujer et al., 1999). The
Intissar Khoja, Research Unit: Industrial systems study and renewable
major difference between these two models is that the recent energy (ESIER), the National Engineering School of Monastir (ENIM),
one recognizes the influence of storage polymers in the hetero- University of Monastir, Av. Ibn El Jazzar Skanes (5019), Tunisia.
trophic activated sludge conversion (Mulas, 2006). Email: intissar_khoja@yahoo.com
2 Transactions of the Institute of Measurement and Control 00(0)

2008; Iacopozzi et al., 2007; Jeppsson et al.[AQ: 3], 1993;


Julien et al., 1998; Smets et al., 2003).
For all these different models, a variety of identification
methods have been created. Among them we can cite: linear-
ized maximum likelihood (Kabouris et al.[AQ: 4], 1996),
extended Kalman filter (Jeppsson et al.[AQ: 5], 1993), the cal-
culus of state variables sensibilities (Caraman et al., 2006),
subspace method (Sotomayer et al., 2003) the Nelder-Mead
(simplex) method (Gomez Quintero, 2002; Gomez Quintero
et al., 2000; Julien, 1997[AQ: 6]), recursive prediction error
method (Ekman, 2008), and the minimization of an
Euclidian-distance criterion (Caraman et al.[AQ: 7], 2008).
The considered model in our paper is the one developed Figure 1. Activated sludge process.
by Gomez Quintero in 2000. It has been identified using the
Nelder-Mead method (Gomez Quintero et al., 2000). The
identification results are convincing but still, there is a signifi- unforeseen fluctuations of the wastewater’s flow and load, so
cant error between model’s response and measurements. In as to always guarantee a certain quality of the purified water.
order to lessen this error to have a more realistic presentation This can be considered as an automation problem and its res-
for the activated sludge process (ASP), we tried to apply more olution is based on control strategies. These strategies can be
efficient identification techniques. only established with the use of a model that is efficiently rep-
During the last decades, metaheuristic algorithms inspired resentative and simple at the same time. This reason leads us
by natural phenomena have attracted considerable attention to develop a simpler model than the well-known family of
as powerful means for solving complex optimization prob- ASM models.
lems. Parameter identification of nonlinear models is tougher
than the one of linear models given the absence of general
analytic results. This kind of issue can be considered as an
optimization one that justifies the use of these non- Description
conventional techniques for general cases (Fidanova et al., The ASP is a system that deals with the treatment of indus-
2014; Hachana et al., 2013; Kumpanya et al., 2015; trial and sewage wastewaters. There exists a large collection
Talatahari et al., 2014) and for ASP, in particular, such as of designs; however, basically all ASPs consist of two main
genetic algorithm (Holck et al., 2009), particle swarm components: an aeration tank that serves as bioreactor where
optimization[AQ: 8] (Sendrescu, 2013), and so forth. all the bio-phenomena take place and a settling tank or clari-
One of the recent and promising techniques is the cuckoo fier that has a role to separate the AS solids from the treated
search algorithm (CSA). This imitates a very interesting ani- wastewater, as shown in Figure 1.
mal behavior that is the breeding behavior of a certain bird Within the bioreactor, atmospheric air or pure oxygen is
species called cuckoo. Thanks to its excellent performance, it introduced to a primary treated sewage (or industrial waste-
has been investigated in a diversity of optimization problems water) combined with organisms to develop biological flocks
(Bahandari et al., 2014; Basu et al.[AQ: 9], 2013; El-Fergany (AS) while degrading the organic substrates founded in the
et al.[AQ: 10], 2013; Jubaer et al.[AQ: 11], 2014; Swapnil effluent. The mixture of wastewater and biological mass is
et al.[AQ: 12], 2014; Xu et al., 2016), notably the parameter commonly noted as mixed liquor. This latter will be dis-
identification problem (Ma et al., 2013; Mouhamed Aly, charged into the settling tank. The supernatant (treated was-
2013; Roeva et al.[AQ: 13], 2016; Xiang-Tao et al., 2012). tewater) is evacuated to a natural water resource or undergo
So, in this work, the CSA will be applied to identify the further treatment while a part of the settled AS is returned to
proposed nonlinear model. the aeration tank to re-seed the new wastewater entering the
In order to confirm the CSA’s efficiency in achieving this tank and to ensure the desired concentration of mixed liquor
task, simulation results will be evaluated by comparing them in the aeration tank. Due to the solids’ presence in the waste-
with other techniques, which are the Nelder-Mead method water and biological growth, sludges will be eventually accu-
(simplex), AG[AQ: 14] and PSO, as well as the real system’s mulated beyond the desired mixed liquor concentration.
measurements. The paper’s sections are organized as follows: Thus, the remaining AS named waste activated sludge (WAS)
in section 2, a general presentation of the activated sludge’s is removed from the treatment process to keep in balance the
model is given. The concept of the CSA is detailed in section ratio of biomass to food supplied (sewage or wastewater).
3. Section 4 provides the comparison and the discussion of The ASP operation can be summarized into two prime
simulation results for the different proposed approaches. phases that are: first of all, the aerobic/nitrification phase
Finally, the paper is closed with a conclusion. where the microorganisms agglomerate into flocks and pro-
duce the sludge by consuming oxygen. Then the anoxic/deni-
trification stage where the microorganisms pursue the
Model description degradation procedure in the absence of oxygen using an
One of the main difficulties in the management of a waste- external carbon source. In addition to these phases, there is a
water treatment plant is to adapt the treatment process to the transitional one that is usually very short and belongs to the
Ladhari et al. 3

aerobic phase where the microorganisms consume the S_ s = Ds Ssin + Dc Ssc  (Ds + Dc )Ss  (1=YH )(r1 + r2 ) + r7
remaining oxygen after it has been shutting down. ð1Þ
The phases’ switching is ensured by the change of the oxy-
gen transfer coefficient’s value (kLa ). In the aeration phase, its S_ NO3 =  (Ds + Dc )SNO3  ((1  YH )=(2:86YH ))r2 + r3 ð2Þ
value is different from 0 and in the anoxic phase, it is equal to
0. As a result, the aeration operation can be considered S_ NH4 = Ds SNH4 in  (Ds + Dc )SNH4  iNBM (r1 + r2 )  r3 + r6
discontinuous. ð3Þ
The considered ASP in this paper is a pilot unit installed
in the Engineering Laboratory of Environmental Processes S_ O2 =  (Ds + Dc )SO2 + kLa (SO2 sat  SO2 )
ð4Þ
(ELEP) of the National Institution of Applied Sciences  ((1  YH )=YH )  4:57r3
(NIAS) in Toulouse, France. This process is fed by
Toulouse’s sewers’ waste water that has a low mass load. The Where Si, in are the input concentrations of variables and ri
oxygen concentration, which plays a key role in this kind of present the kinetics of the process. They can be written in
plant, cannot be regulated to a constant value. The only these forms
available measurements are the nitrate, the ammonium and
the oxygen concentrations so a reduced model containing r1 = a1 Ss (SO2 =(SO2 + KO2 H )) ð5Þ
these three variables is advisable.
As we mentioned before, many models have been estab- r2 = a1 Ss (SNO3 =(SNO3 + KNO3 ))(KO2 H =(SO2 + KO2 H )) ð6Þ
lished for the ASP but few of them consider the oxygen as a r3 = a2 (SNH4 =(SNH4 + KNH4 AUT ))(SO2 =(SO2 + KO2 AUT )) ð7Þ
state variable like our case. A variety of reduction techniques
have been studied and applied to update these complex mod- r 6 = a3 ð8Þ
els to real-time use. One of the most known methods is the
nonlinear method of regular and singular perturbations. In r7 = a4 ((SO2 =(SO2 + KO2 H ))
fact, it is very simply-used and provides reduced-order models + hNO3 h (SNO3 =(SNO3 + KNO3 ))(KO2 H =(SO2 + KO2 H )))
while preserving the models’ basic structures. Applied on the ð9Þ
reference model developed by Julien in 1997 and deducted
from the ASM (Julien, 1997[AQ: 15]), it divides its 11 state These five kinetics describe the aerobic growth of hetero-
variables into three classes: the fast (small time constant), the trophic biomass, the anoxic growth of heterotrophic biomass,
average and the slow (big time constant) variables and con- the aerobic growth of autotrophic biomass, ammonification
siders the following assumptions: of soluble organic nitrogen and, finally, hydrolysis of
entrapped organics. The different variables are defined in
 The derivatives of the fast state variables became null Table 1 where (a1 , a2 , a3 , a4 ) is a set of specific parameters
because they reach the permanent regime very quickly. for the reduced model. They are to be identified while the
Thus, the fast sub-system will be transformed into a remaining parameters preserve their assigned values in the
set of algebraic equations. reference model since they do not require to be re-identified.
 The derivatives of the slow state variables can be con- Actually, this nonlinear model is composed of two sub-mod-
sidered constants. els: one for nitrification the other one for denitrification.
 The average state variables’ dynamics are irreducible. These two sub-models switch alternately.
The first six rows of Table 1 present the system’s variables
Standing on these hypotheses, the oxygen concentration can- while the remaining rows are for the constant parameters. Ss ,
not be considered as a state variable since it has slow SNO3 , SNH4 and SO2 are the state variables. The oxygen transfer
dynamics (Queinnec et al., 2009). Taking into account the coefficient kLa and the Dilution rate of the input Ds change
inadequacy of this method, the alternative reduction strategy their values in the two considered experiments.
is based on some biochemical considerations (observation of
variables behavior and their influence over kinetics reactions
and other variables) as well as the adjustment of the reduced
CSA
order model to ensure the conservation of controllability and CSA presents a new optimization method inspired by animal
observability properties (Queinnec et al.[AQ: 16], 2009). behavior in nature. It was developed by Yang and Deb in
2009 (Yang and Deb, 2009). This meta-heuristic stand on two
main concepts: firstly, the breeding behavior of cuckoo birds
Nonlinear model and, secondly, the characteristics of Lévy flights of some birds
The proposed reduced nonlinear model is composed of four and fruit flies.
state variables (the biodegradable substrate, the nitrate, the
ammonium and the oxygen) and 11 parameters that make the
manipulation of the ASP more accessible (Gomez Quintero,
Cuckoos’ behavior
2002; Gomez Quintero et al., 2000; Julien, 1997[AQ: 17]; Cuckoos are a family of birds that are well-known not only
Queinnec et al.[AQ: 18], 2009). It can be expressed by the fol- for their captivating voice but also for their fascinating repro-
lowing differential equations ductive strategy. Many researchers have studied and discussed
the aggressive brood parasitism of some cuckoo species.
4 Transactions of the Institute of Measurement and Control 00(0)

Table 1. Equation variables definition.

Notation Name

Ss Biodegradable substrate concentration


SNO3 Nitrogen concentration as nitrate and nitrite
SNH4 Nitrogen concentration as ammonia
SO2 Oxygen concentration
Ds Dilution rate of the purge
kLa Oxygen transfer coefficient
Ssc Carbon concentration
YH Yield coefficient of Heterotrophic biomass
iNBM Mass of nitrogen in the biomass
SO2sat Oxygen saturation concentration
KO2 H Coefficient of average saturation of oxygen for the
heterotrophic biomass
KNH4 AUT Coefficient of average saturation of ammonia for
Figure 2. Example of Lévy flight in 2-dimensional plan.
autotrophic biomass
hNO3 h Correction factor for the hydrolysis in anoxic phase
KNO3 Coefficient of average saturation of nitrate
KO2 aut Coefficient of average saturation of oxygen for the directions being random. The next movement is based on the
autotrophic biomass current position and the transition probability to the next
Dc Dilution rate of the external carbon source location. They can be mathematically modeled by the follow-
a1 Heterotrophic growth rate ing equation
a2 Nitrate production rate by the autotrophic biomass
a3 Hydrolysis rate of the slow biodegradable substrate by the
heterotrophic biomass y = ll ð10Þ
a4 Ammonification of the soluble nitrogen
where l presents the flight length and l presents the variance.
Since 1  l  3, y has an infinite variance.
This parasitic bird is able to guarantee a smooth transition The parasitic cuckoo’s reproduction strategy is essentially
to the next generation without making the smallest effort to based on finding the suitable host nest. Generally, the nest’s
build a nest or feed its own offspring like the other birds. The search strategy has demonstrated to be similar to the one for
cuckoo female observes carefully the candidate host nest in food, given that in nature animals dig for their food randomly
order to determine their time of laying eggs or even to know or quasi-randomly. They follow trajectories or directions that
the characteristics of their eggs. Some types of cuckoos called can be defined by certain mathematical equations.
Tapera are very intelligent. They have gone so far as to Different studies have proved that the flight behavior of
develop the ability to imitate the host’s eggs in shape and many animals and insects has the same characteristics as the
color. This reduces the probability of their eggs being thrown typical Lévy flights. A recent one shows that fruit flies or
away and thereby increasing their reproductive capacity. "Drosophila melanogaster" travel their landscape using a
The female lays an egg in the host’s nest when their own series of straight trajectories followed by a sudden turn of
has just been laid. The cuckoo’s eggs have the ability to hatch 90°, which leads to a search plan of Lévy flight style. This lat-
quicker than the host’s eggs. Guided by their natural instinct ter is generally represented by small random steps succeeded
of breeding, the host birds will hatch and bring food to the by large jumps, as illustrated in Figure 2. These long jumps
small cuckoos. Once hatching, the first instinct action of the allow the cuckoos to explore efficiently the searching space
small cuckoo is to chase out the host eggs by blindly discard- especially the large-scale one in the minimum time, unlike the
ing them out of the nest. This will increase the small cuckoo’s normal random walk. The cuckoo breeding behavior mingled
share of food arranged by its host bird. More studies have with the Lévy flight behavior produces an effective meta-
shown that the cuckoo chick has the ability to imitate the call heuristic technique for optimization problems that will be
of host chicks to gain access to more feeding opportunity. detailed in the next part.
In order to raise the surviving chance of their
children[AQ: 19], some species like the Ani and Guira may
remove the host’s eggs. Some host species may engage in con- Algorithm
flict with the intruder cuckoo. When host birds discover the The CSA is based on the following rules:
presence of eggs that are not their own, thanks for example to
a sensitive skin area under their bellies, they throw them out
 Each cuckoo picks a nest randomly in which it lays
or abandon the nest by constructing a new one elsewhere.
one egg at a time;
 The best nest with the highest quality of eggs can pass
on to the new generation;
Lévy flights  The number of the host nests is fixed, and the host
Lévy flights are random walks characterized by their step bird can reveal the foreign posed egg with a probabil-
lengths that have a certain probability distribution, with their ity Pa 2 [0, 1].
Ladhari et al. 5
!
u
Start a  Levy(l) ’ k 1 ðxbest  xi Þ ð13Þ
ðjvjÞb

Choose a random population of n host where k is the Lévy multiplication coefficient fixed by users,
b = 1:5, u and v are deducted from the normal distribution
Get a cuckoo randomly by Lévy flights, curves
i and evaluate its fitness Fi
u ’ N (0, s2u ) v ’ N (0, s2v ) ð14Þ

Select a nest among n randomly, j where


 1=b
Fi ≤ Fj
G(1 + b) sin (pb=2)
Yes su = , sv = 1 ð15Þ
bG½(1 + b)=22(b1)=2
No

Let j as the solution with GðÞ presents the Gamma function.

Replace j with the new solution Simulation results and discussion

Abandon a fraction (Pa) of worst nests


Numerical example
and built new ones at new locations via In order to prove the utility of our method, we first tried to
Lévy flights apply it to an artificial model of the same nonlinear structure
as the considered ASM with sensible selected parameters. We
Keep the current best solution
have chosen the following nonlinear model

x_ 1 =  Ux1 x2  a1 x1 (x4 x2 =(x4 + x2 ))


ð16Þ
t≤maxiter + a4 (x4 =(x4 + x2 )) + 2x2 =(x4 + x2 x2 )

Yes x_ 2 =  Ux2 + a1 x1 x2 =(x4 + x2 )  a2 x3 x4 =(x4 + x3 ) ð17Þ


No
End x_ 3 =  Ux3 + 3a1 x1 (x4 x2 =(x4 x4 + x2 ))
pffiffiffi ð18Þ
 a2 (x4 x2 =(x4 + x2 )) + 5a3
Figure 3. Flow chart of CSA.
x_ 4 =  Ux4 + SC(9:8  x4 )  20a1 x1 (x4 =(x4 + 3))
pffiffiffi ð19Þ
 8a2 =10(x4 x3 =(x4 x4 + x3 x2 ))
If a cuckoo’s egg has been located, the host bird can throw it
away or desert its nest. In the two cases, a new nest will be where: U is equal to10; SC is the switching condition. x1 , x2 ,
built with the probability Pa for a fixed nests’ number. The x3 , x4 are the states variables. a1 , a2 , a3 , a4 are the parameters
CSA can be summarized in the flowchart of Figure 3. to be identified.
A new solution xðt + 1Þ will be randomly generated for a We assume that
cuckoo i, by the mean of Lévy flight as shown in the follow-
ing equation Y = C:X ð20Þ

ðt + 1 Þ
xi = xti + a  Levy(l) ð11Þ with
2 3
where xti presents samples (eggs), i is the sample number and t 0 1 0 0
C=40 0 1 05 ð21Þ
is the number of iteration, while a  0 is the step size.
0 0 0 1
The value of the step size is related to the scale of the con-
sidered problem and controlled by its constraints. So, it is
In order to have measurements, we added a Gaussian noise
important to tune this value. The product  means the entry-
only to our outputs as shown in the following figures. Thus,
wise multiplication. Levy(l) is drawn from Lévy distribution
there is no noise on state x1.
as follows In order to confirm the effectiveness of our method, we
have chosen two evaluation criteria that are the mean square
Levy(l)’y = ll ð12Þ error (MSE) and the Standard Deviation (SD). The first can
be written as follows
The Lévy distribution can be simplified by the following
equation X
n X
n
MSE = (1=n) e2 = (1=n) ( Ybi  Yi )2 ð22Þ
i=1 i=1
6 Transactions of the Institute of Measurement and Control 00(0)

15 12
Model
GA
PSO 11
CSA

10 10

9
x1

x3
5 8

6
0
0 0.05 0.1 0.15 0.2 0.25
Time
5
0 0.05 0.1 0.15 0.2 0.25
Figure 4. The first state variable’s performance.[AQ: 20] Time

Figure 6. The third state variable’s performance.

14

12
0.18

10 0.16

0.14
8
x2

0.12
6
0.1

4
x4

0.08

2 0.06

0.04
0
0 0.05 0.1 0.15 0.2 0.25
Time 0.02

0
Figure 5. The second state variable’s performance.
-0.02
0 0.05 0.1 0.15 0.2 0.25
Time

with: Ybi and Yi representing, respectively, the estimated and


Figure 7. The fourth state variable’s performance.
the measured system outputs at each sample time i,
(i=1,.,n).
e is the error between these outputs (real and estimated
ones) and n is the number of samples. with: n presenting the number of data points. xi and x, respec-
The second one can be defined as follows tively, the value of each data and the mean value of all data.
As we can see, the obtained results shown in Figures 4–7
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi and Tables 2–3 prove the ability of the CSA to surpass other
uP
u n methods in identifying the unknown parameters of the con-
u ðx  xÞ
ti = 1 i sidered nonlinear model. It is noteworthy that we have tried
SD = ð23Þ
n1 to keep the same set of algorithm’s parameters chosen for the

Table 2. Identified parameters.

Parameters Model GA (mean/Variance) PSO (mean/Variance) CSA (mean/Variance)

a1 60 60/0.81 60.24/1.1 59.84/9.94


a2 190 212.65/43.7 212/35.77 212.44/20.59
a3 50 50.5/0.95 50/0 50.136/0.19
a4 980 989.19/14.66 989.09/11.55 988.15/9.65
Ladhari et al. 7

Table 3. Evaluation criteria. Table 5. Parameter settings.

Model GA PSO CSA Parameters Value

MSE 0.206 0.205 0.204 0.203 CSA GA PSO


SD - 0.0001 0.001 0.002
Crossover probability - 0.9 -
Mutation probability - 0.1 -
Selection probability - 0.5 -
Inertia weight (w) - - 0.4<w<0.9
Table 4. Influent wastewater characteristics.
Acceleration coefficient (c1) - - 2
Acceleration coefficient (c2) - - 2
Inflow concentration(g.m23) Flow(m3.d21) (d21)
The probability (Pa) 0.25 - -
SNH4in Ssin Ssc QS QC kLa

Exp.1 61.2 180.13 16000 0.0343 0.0005 114


Exp.2 62.8 183.6 16000 0.0343* 0.0005 225 other optimization techniques that are, firstly, the well-known
classical Nelder-Mead method (Nelder et al.[AQ: 21], 1965)
*: At t=3 hours, QS value changes to 0.0603. and, secondly, the most applied meta-heuristics: the genetic
algorithm (Holland, 1992) and the PSO (Kenndy et al., 1995).
In this paper, our aim is to extract the specific parameters
experimental example. It is known that the meta-heuristic values of the reduced nonlinear model that presents the per-
techniques are highly dependent on the considered problem formance of an activated sludge process using the CSA. These
and the choice of these parameters. Another set of parameters parameters (a1 , a2 , a3 , a4 ) correspond to the nests’ positions
may give better results. at each iteration. At first, the positions of all nests are chosen
randomly. Then, at each iteration, they will be updated via
Lévy flights within the desired search space. At the end of all
ASP the iterations, the obtained result corresponds to the best
In order to identify mathematical models that describe the nests’ positions that are actually the optimal parameters val-
biological processes two important facts must be taken into ues of the model.
consideration: firstly, the big number of state variables and The application of the different considered techniques in
parameters to identify, and, secondly, the small quantity or this paper, especially the intelligent ones, need a careful tune
the poor quality of the available measurements. of their several characteristic parameters, since they highly
Parameter identification procedure aims to determine the influence the performance of these algorithms. The unfit
model’s parameters using a set of input-output measures. It choice of these parameters contributes to a cyclic behavior or
compares these real measurements with the ones obtained by total divergence of the algorithm. Various tests have been
the estimated model using a mathematical function named achieved to obtain the suitable characteristic parameters.
objective function. Thus, the parameter identification problem They are presented in Table 5.
can be considered as an optimization problem by minimizing Not only the choice of the different algorithms’ parameters
the error between the estimated and real outputs values. The influences a computational burden but also the choice of the
considered objective function (fitness) that needs to be mini- objective function (fitness), the[AQ: 22] parameters’ ranges
mized is the mean squared error (MSE). because they are strongly dependent on these choices. A dif-
In this section, the CSA is employed to identify the para- ferent set of choices may give another result. The parameters’
meters of a nonlinear model that describes an activated sludge number and the algorithm’s complexity also influence compu-
process. It is an off-line case. tational burdens.
The available experimental data are gathered from two For a fair comparison, these algorithms have been run 15
experiments that are conducted for 6 hours with a sample times given their stochastic behavior with a population of 10
time of 20 minutes under different conditions, which are the individuals that are, respectively, chromosomes, birds and
change of values for both the dilution rate of the purge (Ds ) nests. The maximum number of iterations is 100. The results
and oxygen transfer coefficient (kLa ). The remaining para- are shown in Table 6.
meters keep their values of the reference model developed by We considered the mean and the variance values of all the
Julien in 1997. One experiment has been chosen for the identi- runs for each algorithm in order to have a more convenient
fication (experiment 2) while the other has been dedicated to result. This latter will be set against the result of the Nelder-
the validation (experiment 1) in order to check the compat- Mead method as well as the calculated values obtained from
ibility of the identified model with a different experimental mathematical relations that link between the reference model
framework. The only available data, as we mentioned before, and the reduced nonlinear model. In other words, the para-
are the measurements of: SNO3 , SNH4 and SO2 . While there are meters ai (i=1,.,4) have been calculated from the identified
no measurements for the Ss . The operating conditions are parameters of the reference model (Gomez Quintero, 2000;
mentioned in Table 4. Julien, 1997). The results are shown in Table 7.
In order to prove its effectiveness and robustness, the Under identical conditions, the obtained results from the
results of CSA will be compared with the ones obtained by different identification techniques will be investigated in the
8 Transactions of the Institute of Measurement and Control 00(0)

Run 15

1087.6

1113.7

1069.1
193.8

191.8
83.2

57.2

76.5

58.2

77.1

56.4
Biodegradable Substrate Concentration (g/m3)

188
18
Simplex
Calculated
16 GA
Run 14

1072.4

1074.3

1030.8
PSO
191.1

188.4

192.2
88.2

56.6

79.3

56.2

56.4
CSA

80
14
Run 13

188.77

999.08

1090.5

1145.1

Ss
12
73.67

52.74

195.7
75.9

56.9

87.3

60.1
190

10
Run 12

1024.8

1097.6

1127.7
187.7

191.6

195.4
76.3

53.7

83.7

57.5

93.2

59.6
8

6
0 0.05 0.1 0.15 0.2 0.25
Run 11

1294.2

1051.7

1089.1
196.5

184.4

190.6

Time(days)
62.8

65.3

72.7
79

55

57

Figure 8. Biodegradable substrate concentration.[AQ: 23]


Run 10

1205.1

1086.8

1178.4
190.3

195.1
62.2

82.7

56.9

87.4

60.3
195
96

Nitrate Concentration (g/m3)


10
Simplex
9 Calculated
1172.3

1098.3

1077.6
Run 9

195.4

193.3

188.3

GA
82.9

59.8

80.2

57.5

56.2
96

8 PSO
CSA
7
Measured
1139.1

1144.9

6
Run 8

194.6

191.4

193.4
1102
85.8

59.4

73.3

57.9

76.7

60.2

SNO3

4
1257.7

1125.4

1089.2
Run 7

3
191.6

196.1

189.3
85.6

60.5

72.5

59.2

57.2
83

1
1079.4

1087.8

1109.2
Run 6

187.2

193.6

191.5

0
56.3

63.4

57.4

80.4

58.9

0 0.05 0.1 0.15 0.2 0.25


86

Time(days)

Figure 9. Nitrate concentration.[AQ: 24]


1071.4

1035.1

1150.3
Run 5

188.2

188.6
77.5

99.5

60.4

62.9

62.3
187
56

model’s simulation as well as the real measurements. The


Table 6. Obtained parameters from different algorithms.

acquired results are given in Figures 8–15. The first four fig-
1123.8

1063.4

1110.7
Run 4

192.8

195.2

ures display the simulation results of experiment 2 while the


91.5

58.9

88.9

60.5

63.2
187
63

remaining four figures show the results of experiment 1.


Graphical comparison presents an efficient mean to clearly
show the accuracy of the considered method thanks to the
1074.3

1025.9

1000.7
Run 3

188.2

186.4

188.8
72.2

55.9

93.1

57.3

77.3

58.3

quantitative measure of the difference between the estimated


and measured values. The potential reasons for this difference
in our study can be the simplifications made in the first place
1018.4

1069.5

1018.1

on the ASM1 to obtain the reference model then, on this lat-


Run 2

190.1

189.6
91.9

53.9

83.2
187

ter to obtain the considered reduced nonlinear model in the


99

59

59

second place. For more detail, see Gomez Quintero et al.


(2000). These simplifications are aiming to produce a simple
1153.7

1144.9

1000.2
Run 1

representative model for the control strategies by eliminating


195.1

195.2

193.5
87.8

59.1

74.8

64.1

58.8
65

some phenomena that may cause the differences between


model response and real-life measurements. Focusing on the
different figures, we note the absence of the experimental data
CSA
PSO
GA

for the biodegradable substrate concentration. So, any


Ladhari et al. 9

Table 7. Identified parameters.

Parameters Calculated values Simplex GA (mean/Variance) PSO (mean/Variance) CSA (mean/Variance)

a1 95.81 62.59 82.21/77.47 83.85/107.27 76.38/109.52


a2 197.65 187.37 190.92/10.98 191.19/12.07 190.755/8.91
a3 78.88 52.63 57.56/11.57 58.76/5.02 59.275/4.57
a4 1516.1 987.2 1110.73/7501.02 1110.65/1779.08 1086.055/3227.68

Ammonia Concentration (g/m3) Biodegradable Substrate Concentration (g/m3)


12 15
Simplex Simplex
Calculated 14 Calculated
GA GA
10
PSO 13 PSO
CSA CSA
Measured 12
8

11

Ss
SNH4

6
10

9
4
8

2 7

6
0 0.05 0.1 0.15 0.2 0.25
0 Time(days)
0 0.05 0.1 0.15 0.2 0.25
Time(days)

Figure 12. Biodegradable substrate concentration.[AQ: 27]


Figure 10. Ammonia concentration.[AQ: 25]

Dissolved Oxygen Concentration (g/m3) Nitrate Concentration (g/m3)


6 12
Simplex Simplex
Calculated Calculated
5 GA 10 GA
PSO PSO
CSA CSA
4 Measured
Measured 8

3
SNO3
SO2

2
4

2
0

0
-1 0 0.05 0.1 0.15 0.2 0.25
0 0.05 0.1 0.15 0.2 0.25 Time(days)
Time(days)

Figure 13. Nitrate concentration.[AQ: 28]


Figure 11. Dissolved oxygen concentration.[AQ: 26]

interpretation concerning the estimated model accuracy for example, the MSE and the SD have been chosen as criteria to
this variable cannot be validated. In this way, the validation evaluate the performance of the distinct algorithms[AQ: 31].
of the estimated nonlinear model can only be made on the Results are shown in Table 8.
nitrate and ammonia responses. The performance of these Comparing the CSA’s values with the other methods –
two responses in the two experiments illustrates the outper- classical (Nelder-Mead) or intelligent (GA and PSO) – con-
formance of the CSA in providing an estimated model that firms its ability to provide a reduced-order model that has the
successfully predict the real system responses. same dynamic as the real one that is very complicated to be
To further prove how close the estimated model using the useful. This can be clearly seen in the minimum value of MSE
CSA with the real measurements and like our numerical and the maximum value of SD given by the CSA.
10 Transactions of the Institute of Measurement and Control 00(0)

Ammonia Concentration (g/m3)


Finally, the long jumps of Lévy flights allow the algorithm to
9 efficiently avoid local optima and quickly converge toward
Simplex
8 Calculated the best solution.
GA All this allows us to draw the following conclusion that
7 PSO
CSA
CSA is a simple, reduced-number-parameters and robust
6 Measured method that makes it more beneficial as a parameter identifi-
cation technique than the others techniques.
5
SNH4

3 Conclusion
2 In this paper, the parameter identification has been formu-
1
lated as a CSA-based optimization problem. The considered
model to be identified is a reduced-order nonlinear model for
0
0 0.05 0.1 0.15 0.2 0.25 an activated sludge wastewater treatment process. The experi-
Time(days) mental data are collected from two experiments conducted in
different aeration conditions. These data are compared with
Figure 14. Ammonia concentration.[AQ: 29] different model’s responses obtained by using the distinct
optimization techniques that can be classified into two cate-
gories: classical (Nelder-Mead) and intelligent (GA, PSO, and
CSA). Since the CSA provides the smallest error, it is the ade-
Dissolved Oxygen Concentration (g/m3)
2 quate technique for parameter identification with a fine preci-
Simplex
Calculated sion. Simulation results also reveal the validity and the
1.5
GA
PSO
effectiveness of the considered method showing the closeness
CSA between data and the model’s performance.
Measured
It is noteworthy that the obtained results can be further
1
improved using other meta-heuristic-based methods in order
SO2

to provide a more valuable outcome that can be investigated


0.5
in establishing the control strategies.

0
Declaration of conflicting interests
-0.5 The author(s) declared no potential conflict of interests with
0 0.05 0.1 0.15 0.2 0.25
Time(days) respect to the research, authorship and/or publication of this
article.
Figure 15. Dissolved oxygen concentration.[AQ: 30]

Funding
Table 8. Evaluation criteria. This research received no specific grant from any funding
agency in the public, commercial, or not-for-profit sectors.
Simplex Calculated GA PSO CSA

MSE 1.41 2.3 1.4 1.38 1.32 References


SD - - 0.0026 7e-5 0.0014 Bahandari AK, Soni V, Kumar A, et al. (2014) Cuckoo search algo-
rithm based satellite image contrast and brightness enhancement
using DWT–SVD. ISA Transactions 53(4): 1286–1296.
Basu M and Chowdhury A (2013) Cuckoo search algorithm for eco-
Unlike the other considered intelligent techniques, the nomic dispatch. Energy 60: 99–108.
CSA presents an effective parameter identification method Birs IR, Nascu I, Darab C, et al. (2016) Modelling and calibration of
thanks to three main characteristics, which are: the simplicity, a conventional activated sludge wastewater treatment plant. In:
the reduced number of parameters and the robustness. For IEEE International Conference of Automation, Quality and Test-
ing, Robotics, Cluj-Napoca, Romania.
the first one, the updating of the solutions only depends on
Caraman S and Barbu M (2008) The identification and robust con-
the Lévy flight equation. For the second one, the main para- trol of a biological wastewater treatment process. In: IEEE Inter-
meters are: the population size, the number of iterations and national Conference on Automation, Quality and Testing, Robotics
the switching parameter Pa . The most important one is Pa 3, Cluj-Napoca, Romania, 22–25 May, pp. 37–42.[AQ: 32]
given that the other two parameters are common among all Caraman SV, Barbu M and Dumitrascu G (2006) Wastewater treat-
the population-based methods (GA and PSO). This para- ment process identification based on the calculus of state variables
meter essentially controls the balance of randomization, the sensibilities with respect of the process coefficients. In: IEEE Inter-
elitism and the local search, which give the CSA the opportu- national Conference on Automation, Quality and Testing, Robotics
nity to have an efficient exploration of the searching space. 2, pp. 199–204.[AQ: 33]
Ladhari et al. 11

Ekman M (2008) Bilinear black-box identification and MPC of the Julien S Modélisation et estimation pour le controle d’un procédé
activated sludge process. Journal of Process Control 18(7–8): boues activées éliminant l’azote des eaux résiduaires urbaines.
643–653. PhD Thesis, Toulouse university, France.[AQ: 40]
El-Fergany AA and Abdelaziz AY (2013) Capacitor allocations in Julien S, Babary JP and Lessard P (1998) Theoretical and practical
radial distribution networks using cuckoo search algorithm. IET identifiability of a reduced order model in an activated sludge pro-
Generation, Transmission & Distribution 8(1): 223–232. cess doing nitrification and denitrification. Water Science and
Fidanova S, Paprzycki M and Roeva O (2014) Hybrid GA-ACO Technology 37(12): 309–316.
algorithm for a model parameters identification problem. In: Pro- Julien S, Lessard P and Barbary JP (1999) A reduced order model for
ceedings of the Federated Conference on Computer Science and control of a single reactor activated sludge process. Mathematical
Information Systems: 413–420.[AQ: 34] and Computer Modelling of Dynamical Systems 5(4): 337–350.
Gernaey KV, van Loosdrecht MC, Henze M, et al. (2004) Activated Kabouris JC and Georgakakos AP (1996) Parameter and state esti-
sludge wastewater treatment plant modelling and simulation: state mation of the activated sludge process-II. Applications. Water
of the art. Environmental Modelling & Software 19(9): 763–783. Research 30(12): 2867–2882.
Gomez-Quintero C (2002) Modélisation et estimation robuste pour Kenndy J and Eberhart RC (1995) Particle swarm optimization. In:
un procédé boues activées en alternance de phases. PhD Thesis, Proceedings of IEEE International Conference on Neural Networks
Toulouse University, France. 4, pp. 1942–1948.[AQ: 41]
Gomez Quintero C, Queinnec I and Babary JP (2000) A reduced non- Kumpanya D, Thaiparnat S and Puangdownreong D (2015) Para-
linear model of an activated sludge process. In: IFAC Advanced meter identification of BLDC motor model via metaheuristic opti-
Control of Chemical Process, Pisa, Italy.[AQ: 35] mization techniques. Procedia Manufacturing 4: 322–327.[AQ: 42]
González-Miranda O, Rios-Bolivar M and Gómez-Quintero C (2009) Ma J, Ting TO, Man KL, et al. (2013) Parameter estimation of photo-
Adaptive output feedback regulation of an alternating activated voltaic models via cuckoo search. Journal of Applied Mathematics.
sludge process. In: Proceedings of the European Control Confer- DOI: 362619.[AQ: 43]
ence, Budapest, Hungary, pp. 424–429.[AQ: 36] Mouhamed Aly W (2013) Evaluation of cuckoo search usage for
Gujer W, Henze M, Mino T, et al. (1995) The activated sludge model model parameters estimation. International Journal of Computer
no. 2: Biological phosphorus removal. Water Science and Technol- Applications 78(11).[AQ: 44]
ogy 31(2): 1–11. Mulas M (2006) Modelling and control of activated sludge processes.
Gujer W, Henze M, Mino T, et al. (1999) Activated sludge model Università degli Studi di Cagliari.[AQ: 45]
No.3. Water Science and Technology. 39(1): 183–193. Nelder JA and Mead R (1965) A simplex method for function minimi-
Hachana O, Hemsas KE, Tina GM, et al. (2013) Comparison of dif- zation. The Computer Journal 7(4): 308–313.
ferent metaheuristic algorithms for parameter identification of Queinnec I and Gomez-Quintero C (2009) Reduced modeling and
photovoltaic cell/module. Journal of Renewable and Sustainable state observation of an activated sludge process. Biotechnology
Energy 5(5).[AQ: 37] Progress 25(3): 654–666.
Hadj-Sadok MZ and Gouzé JL (2001) Estimation of uncertain mod- Rieger L, Koch G, Kühni M, et al. (2001) The EAWAG Bio-P mod-
els of activated sludge processes with interval observers. Journal of ule for activated sludge model No. 3. Water Research 35(16):
Process Control 11(3): 299–310. 3887–3903.
Hauduc H, Rieger L, Ohtsuki T, et al. (2011). Activated sludge mod- Roeva O and Atanassova V (2016) Cuckoo search algorithm for
elling: Development and potential use of a practical applications model parameter identification. International Journal Bioautoma-
database. Water Science and Technology 63(10): 2164–2182. tion 20(4): 483–492.
Hauduc H, Rieger L, Takács I, et al. (2010). A systematic approach Sendrescu D (2013) Parameter identification of anaerobic wastewater
for model verification: application on seven published activated treatment bioprocesses using particle swarm optimization. Mathe-
sludge models. Water Science and Technology 61(4): 825–839. matical Problems in Engineering. DOI 103748.[AQ: 46]
Henze M, Grady CPL Jr, Gujer W, et al. (1987) Activated sludge Smets IY, Haegebaert JV, Carrette R, et al. (2003) Linearization of
model no. 1. International Association on water pollution the activated sludge model ASM1 for fast and reliable predictions.
Research and Control. Scientific and Technical Reports No. 1. Water Research 37(8): 1831–1851.
Henze M, Gujer W, Mino T, et al. (1999) Activated sludge model Sotomayer OAZ, Park SW and Garcia C (2003) Multivariable identi-
No.2D, ASM2D. Water Science and Technology 39(1): 165–182. fication of an activated sludge process with susbspace-based algo-
Holck P, Sorsa A and Leiviskä K (2009) Parameter identification in rithms. Control Engineering Practice 11(8): 961–969.
the activated sludge process. Chemical Engineering Transactions Swapnil N and Prabin KP (2014) Identification and cuckoo PI-PD
17: 1293–1298.[AQ: 38] controller design for stable and stable processes. Transactions of
Holenda B, Domokos E, Redey A, et al. (2008). Dissolved oxygen the Institute of Measurement and Control 37(6): 708–720.
control of the activated sludge wastewater treatment process using Talatahari S, Mohaggeg H, Najafi KH, et al. (2014) Solving para-
model predictive control. Computers & Chemical Engineering meter identification of nonlinear problems by artificial bee colony
32(6): 1270–1278. algorithm. Mathematical Problems in Engineering.[AQ: 47]
Holland JH (1992) Genetic algorithms. Scientific American 267(1): Vrečko D, Hvala N and Stražar M (2011) The application of model
66–73. predictive control of ammonia nitrogen in an activated sludge pro-
Iacopozzi I, Innocenti V, Marsili-Libelli S, et al. (2007). A modified cess. Water Science and Technology 64(5): 1115–1121.
Activated Sludge Model No. 3 (ASM3) with two-step nitrifica- Wu X, Yang Y, Wu G, et al. (2016) Simulation and optimization of a
tion–denitrification. Environmental Modelling & Software 22(6): coking wastewater biological treatment process by activated
847–861. sludge models (ASM). Journal of Environmental Management 165:
Jeppsson U and Olsson G (1993) Reduced order models for on-line 235–242.[AQ: 48]
parameter identification of the activated sludge process. Water Xiang-Tao L and Ming-Hao Y (2012) Parameter estimation for chao-
Science and Technology 28(11–12):173–183. tic systems using the cuckoo search algorithm with an orthogonal
Jubaer A and Zainal S (2014) A maximum power point tracking learning method. Chenise Physic B 21(5).[AQ: 49]
(MPPT) for PV system using cuckoo search with partial shading
capability. Applied Energy 119: 118–130.[AQ: 39]
12 Transactions of the Institute of Measurement and Control 00(0)

Xu H, Liu J and Lu Z (2016) Structural damage identification based Zhang P, Yuan M and Wang H (2008) Improvement of nitrogen
on cuckoo search algorithm. Advances in Structural Engineering removal and reduction of operating costs in an activated sludge
19(5): 849–859. process with feedforward–cascade control strategy. Biochemical
Yang XS and Deb S (2009) Cuckoo search via Lévy flights. In: IEEE Engineering Journal 41(1): 53–58.
World Congress of Nature & Biologically Inspired Computing, pp.
210–214.[AQ: 50]

View publication stats

You might also like