Alexandria Engineering Journal (2022) 61, 9287–9295

H O S T E D BY
Alexandria University

Alexandria Engineering Journal

www.elsevier.com/locate/aej
www.sciencedirect.com

An Artificial Neural Network Mechanism for

Optimizing the Water Treatment Process and
Desalination Process
Ahmed I. Taloba

Department of Computer Science, College of Science and Arts in Qurayyat, Jouf University, Saudi Arabia

Received 10 February 2022; revised 1 March 2022; accepted 15 March 2022

KEYWORDS Abstract An artificial neural network (ANN) was created to predict the operation of a reverse
Water treatment; osmosis desalination process, and then it was utilized to supply water temperature modeling. The
Water desalination; techniques used for producing the new water by utilizing the water treatment process and desalina-
Artificial neural network; tion of seawater process are growing rapidly. Desalination produces water for residential use, the
Multilayer network of processing industry, water supply, and other reasons. Membrane procedures (Reverse Osmosis
perceptron RO) or thermal desalination are always the most common desalination techniques. The optimiza-
tion process consists of some units such as coagulation, sedimentation, disinfection, settling, and
filtration. The process of desalination has various parameters such as vacuum pressure, feed salt
concentration, temperature feed into the inlet, and rate of feed flow. For the desalination process,
reverse osmosis is the most helpful process for desalination that can be coupled with more variable
renewable energy sources are wind and solar. Using a couple of renewable energy to produce fresh-
water from saltwater can able to reduce the impact on the environment that desalination can create
due to the power consumption of energy. The optimization of water treatment and desalination
experimental results was developed by the model of an artificial neural network (ANN).
Ó 2022 The Authors. Published by Elsevier B.V. on behalf of Faculty of Engineering, Alexandria
University This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/
4.0/).

1. Introduction demographic changes and improvements in general living stan-

dards at all levels, which facilitate the growth of freshwater for
Everyday living beings are performing a variety of remarkable a range of uses [2]. The important points are receiving ade-
activities that indirectly or directly include the usage of water. quate thought on debate for providing the water changes, then
Water is necessary for all industries activity, domestic pur- all these adjustments that influence the decisions. The two fac-
poses, and agriculture [1]. Many causes contribute to the tors that can be linked directly are the condition of water
increasing world need for freshwater consumption, including resources and the price of water treatment [3]. The quality of
water resources is low that can increase the cost for treatment,
E-mail address: aitaloba@ju.edu.sa
the insufficient budget allocation reduces the treatment for
Peer review under responsibility of Faculty of Engineering, Alexandria
consumed water desirability.
University.
https://doi.org/10.1016/j.aej.2022.03.029
1110-0168 Ó 2022 The Authors. Published by Elsevier B.V. on behalf of Faculty of Engineering, Alexandria University
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
9288 A.I. Taloba

Water treatment and desalination remove microorganisms, result information are detections in false and the detection is
salts, and minerals in freshwater or seawater to make it safe for missed. The wavelet denoise (WD) is joint through the PCA
human use. Freshwater desalination from rivers or groundwa- that forms a new process that detects the responsibility on
ter, wastewater reuse, and water management all demand more the feed water treatment process. This result is further com-
energy than saltwater desalination [4]. For desalination to be pared with some of the existing methods. Then the advantage
carried out, the salt in the water should leave the low concen- of PCA-WD, which can be against the classical SWPCA and
trated source to the highly concentrated source. This can be PCA is getting finally convinced [13].
done by reverse osmosis which requires overcoming the natu- The current leading technology in the process of desalina-
rally occurring osmotic pressure and adding additional pres- tion is Reverse Osmosis. But in recent days, membrane fouling
sure to drive the process in the reverse direction. This is widely used. Fouling is a complex mechanism that was not
process requires high power consumption due to the need for well understood in full-scale plants. Several types of research
high-pressure pumps to create such additional high pressure are done in fouling and are found that the fouling characteris-
[5]. tics are well understood in lab or pilot scale systems but not in
The osmotic pressure is a naturally occurring process and full-scale systems. Both the ANN and tree model are used to
hence it is limited to its efficiency. The seawater desalination analyze the long-term performance of a full-scale reverse
plant has around 50% efficiency in the recovery rate, i.e., for osmosis desalination plant and the results show that this sys-
1-liter untreated water, ø liter freshwater can be obtained tem produces a high correlation coefficient among the mea-
and the remaining ø liter is redirected to the sea with the whole sured and simulated output variables [14].
of salt that entered into the plant[6]. In other words, for 1-liter With the help of ANN, a Sweeping gas membrane distilla-
freshwater, 2 L of water is to be processed [7]. There are 5 main tion process (SGMD) for desalination of water and to predict
components in a desalination plant. They are the Intake its performance index based on 53 experiments. For this, a
(18%), Pre-treatment (22%), Treatment (49%), Post- feed-forward network is used. With the help of Monte Carlo
Treatment (3%), and Waste Disposal (8%) [8]. The proportion simulation, the optimum point of feed inlet temperature, the
in parenthesis relates to the cost of every element in a water feed flow rate, and the airflow rate on the SGMD performance
treatment plant, which might vary depending on location, index was determined [15]. The reverse osmosis-based water
ground condition, and distances from the facility to the coast- desalination plant is a highly non-linear multi-input multi-
line. In this study, a cost-effective ANN-based system for puri- output system that can be affected by some uncertainties, con-
fying seawater or water from other origins to make it straints, or some other physical phenomena. Hence effective
appropriate for irrigation and human consumption is pre- control strategy is needed and this can be achieved by using
sented [9]. a Non-linear Model Predictive Controller (NMPC). A Recur-
Non-linear multivariate regression problems are focused on rent Neural Network (RNN) base Long-Short Term Memory
the use of ANN models. This is defined as Black-box design (LSTM) is used to develop a predictive model for controlling
because it is not depending on the presentation of specific with the desalination plant is introduced [16].
the actual significance of the technology that is under inquiry Actuator faults are more common in a smaller reverse
that is depending on the result of developments. The ANN osmosis-based desalination plant, which reduces the water
design can easily progress by using the experimental informa- quality or energy loss in the plant. A multi-input multi-
tion from the process to be spread over the model of the exper- output (MIMO) reverse osmosis desalination plant is con-
imental system [10]. This model can be applied to various fields trolled along with noise, model mismatch, and actuator fault
of medicine, technology, and science [11]. ANN is applied to [17]. An optimal strategy for the control of reverse osmosis
seawater desalination, in contrast with the deterministic system membrane module with recirculation is demonstrated using
that is based on physical laws. In ANN, the model of general the model of reverse osmosis membrane to remove the impuri-
additive is comfortable to derive the complex design that is ties in the water that is to be recycled and to optimize the time
from the actual information data, they requiring without the taken to for the process of recycling [18].
assumption for conditions are ideal relatively with perfor- An enhanced multi-objective optimum control (MOOC)
mance are better [12]. method is presented in this study to achieve operational excel-
The remaining part and aim of this paper are explained lence, meet treatment efficiency (EQ), and reduce the energy
about the optimization of water treatment and desalination consumption (EC) in the wastewater treatment process
by using the ANN method; part 2 defines the highlight of pre- (WWTP). First, adaptable kernel function modeling of the
vious effort that can be done by the scholars in this domain processes is built for the suggested MOOC method, which
with various experimental tasks; part 3 represent the system can capture the complicated dynamics of EQ and EC. How-
model of this research; part 4 offering the proposed methodol- ever, to account for WWTP, a multi - objective optimization
ogy architecture model and its mechanism; part 5 represents situation is caused. Second, an enhanced based multi - objec-
the evaluation and performance of the result and discussion tive particle swarm optimizing (MOPSO) technique is devised
and part 6 represents the work achieved in conclusion. to minimize the set objectives utilizing a self-adaptive flight
characteristics method and a multi-objective gradient (MOG)
2. Related work method. The treatment method then yields the ideal dissolved
oxygen (SO) and nitrate (SNO) predefined. Third, in the sug-
The important part of value boilers is the Feedwater treatment gested MOOC technique, an adaptive neuro - fuzzy inference
process (FWTP) then the fault detection for its improved reli- neural network controllers (FNNC) is used to implement
ability. The principle for the classical component that func- tracking control of the generated set-points. Finally, Bench-
tions to the FWTPs in the earlier effort. The T2 and SPE marking Simulations Model No. 1 (BSM1) is presented to
assess the proposed MOOC strategy’s performance. The pro-
An artificial neural network mechanism 9289

posed technique’s efficacy is demonstrated by experimental & Coagulation:

evidence [19].
To optimize the seawater desalination process in [20], this This is the initial step for the water treatment process. In
can be analyzed by using the combination brand of NF mem- this positive charge, chemicals are poured into the water. On
brane and the parameter with MSF (Multi-stage flash desali- these positive charges on chemical are get neutralize into the
nation process) and SWRO-MSF hybrid condition. The dirt negative charge and the other particles are dissolved in
SWRO is rejected on the latter case used in make-up on the water. When it takes place, thus the particles are got blended
pilot plant MSF distiller. On further hybridization the testing with the chemicals then it forms into large particles are called
program is locked, thus the MSF rejected the heated water was floc [26].
used to feed into the pretreatment NF section. This method is
estimated to develop the feed temperature effect that rise the & Sedimentation:
benefits of NF and NF-SWRO as an energy input. They
address the global optimization design for the MSF processes. During this sedimentation process, thus the floc from the
This model is supplemented into the non-deterministic algo- initial process settles on the bottom water. Then it gets supply
rithm for optimization, thus the simple mixture in MSF-M due to its weight. This process of settling particles in water is
and the brine recycle as MSF-BR is used for the optimization known as sedimentation [27].
process [21]. To evaluate the global optimization design by
using the differential evolution algorithm (DE). The DE is & Filtration:
2.31% better when compared with other literature.
Gravity-driven membrane (GDM) filtration is investigated Once floc is get settled at the bottom water, then the clear
in [22], the evaluation of a hybrid biofilm-submerged GDM clean water at the top that will pass via filters supply with
reactor to improve the time difference between the biofilms the different compositions are gravel, pore sizes, charcoal,
and dissolve organics. By participating in the biofilm reactor and sand are get removed the dissolved particles as parasites,
that with system membrane displayed on the quality is better dust, viruses, chemicals, and bacteria.
than the GDM filtration cells [23], these not reduce the dis-
solved material in seawater efficiently. This can be attributed & Disinfection:
to the degradation of organic solid substances in the carbon
fixation and the feed by the biofilm. Once the filtering process of water is got completed, thus
the disinfected such as chloramine, chlorine can be added to
3. System model kill any outstanding bacteria, viruses, and parasites, and then
the water is protected from the germs, then it will be piped into
In this section, we discussed the system model of the water the businesses and homes.
treatment process and the desalination process. the major
objective of water treatment and desalination plants is to pro- 3.2. Desalination process
duce safe water for drinking and irrigation purposes that are
free from pathogenic micro-organisms, toxic compounds, min- The parameters are used for the desalination process are
eral salts, etc. [24]. explained in this section. Desalination is the process that
removes the excess mineral and salt in the water. This method
3.1. Water treatment process is used for industry, municipality, or any other uses. The
power of the blower is 37 W, the rated power compressor is
450 W, the heat pump COP is 3.0 and the humidifier is
This part explains the detailed process of water treatment, the
water quality is produced by climate change as spring accord- 280
280
80 mm. this setup is used for the desalination pro-
ing to the region rainfall. Sometimes, it can be affected by the cess [28]. Saltwater may include the brackish, surface (streams,
snowfall on the region of the mountain. The water plant treat- rivers), seawater, process waters, industrial feed, and wastewa-
ter. by using the advancement in technology, thus the process
ment is a part of all water resources and the capacity of nom-
inal treated water is 600 I/s. the water treatment is choosing the of desalination has become more effective in cost when it is
raw water regarding the variation of turbidity than the avail- compared with other processes to produce clean water on this
growing demand [8].
ability of information data.
In Fig. 1. represents the overview of various operations that
treat water. The various measurement that is recorded by the cur- & Reverse Osmosis:
rent stream detector such as PH, temperature, turbidity level,
conductivity that is needed to carry out jars that test into deter- When this process is compared with other processes, reverse
mining the aluminum sulfate dose optimal [25]. To conduct out osmosis (RO) is a new process that is used for desalination is
shown in Fig. 2. Thus, the principle includes in the RO process
all the containers test and identify the best dose of aluminum sul-
is as follows, the pressure is used to drive the force then it
phate, many observations of variables collected by streaming pre-
sent sensors such as turbidity level, PH, conductivity, and pushes into the feed water through the semi-permeable mem-
temperature are required. Table 1 displays the raw water param- brane then into the product of stream water and then concen-
trates into the brine stream [29].
eters used in this investigation, as well as the fluctuation intervals.
The procedure step involved in the water treatment process The desalination process by RO mainly comprises the four
is explained as follows: major methods they are:
9290 A.I. Taloba

Fig. 1 Structure of the water treatment process.

applied to the solution of high salt concentration, then the

Table 1 Statistical raw water conditions.
water starts flowing into the direction is reverse via the semi-
Parameter Minimum Maximum permeable membrane, then it leaves the salt on behind, this
Turbidity 1:8 852 process is known as reverse osmosis process [31]. Here the con-
PH 6:82 7:75 figuration of the membrane consists of hollow fibers, the sheet
Temperature (°C) 15 24:71 with spiral and spiral wounds. Thus, the pressure for operating
Conductivity micro s/cm 508 625 the reverse osmosis and the range of nano-filtration is between
the range of 50 and 1000psig.

4. Proposed methodology

 Pre-treatment system In this proposed method, propose an ANN (Artificial neural

 High-pressure pumps network) is introduced on the water treatment process and
 Membrane system desalination process to improve their efficiency [32]. The key
 Post-treatment objective of the process of water treatment and the desalina-
tion process is to help to produce harmless water for drinking
Then the Nano-filtration is similar to the membrane process and irrigation purposes which are free from pathogenic micro-
that is used to remove the divalent ions of salt such as sulphate, organisms, toxic compounds, mineral salts, etc. to avoid the
calcium, and Magnesium. This process is used to remove the solids accumulation, corrosion, and bacteria growth in the dis-
chloride and sodium ions in water [30]. Then osmosis is defined tribution and pipelines are transported in a cost-effective man-
as the natural phenomenon on which the water is from the ner using an advanced ANN model [33].
concentration of low salt into a more solution concentration The ANN-based water treatment and desalination include
through the semi-permeable membrane. Then the pressure is a micro strainer, via ozonation, coagulation/flocculation
An artificial neural network mechanism 9291

Fig. 2 Reverse Osmosis (RO) process.

method, sedimentation process, and filtration process followed Fig. 3 ANN architecture flow chart.
by the ANN training and evaluation model are displayed on
the Fig. 3. The process of coagulation involves pH modifica-
tion and the dosage of coagulant is measured by observing 8k 2 1; 2;    :;    l
untreated turbidity of water and conductivity of electrical
where i = input nodes, s = hidden nodes, l = output
[34]. The help of flocculant is not dosed. For the backwashing
nodes.
filter, 3 principles are used, which include turbidity in water fil-
Apl = output from the lth node of the output layer of the
ter effluent less than 0.2 FNU, particles of greater than 1 mm in
network.
filter effluent less than 320#/ml, and run-time filter is greater
Yps = network input of the pth node.
than 65 h. These 3 criteria are followed to prevent the excessive
Had;l = connection weight between the hidden layer dth node
growth of micro-organisms and to maintain hygiene [35]. The
water from the filter is together with the sludge from the sed- and output layer lth node.
imentation process in a settling tank, which is cleared to the Hwb;d = connection weight between the hidden layer bth
system of sewer and the effluents are cleared to the affluent node and output layer dth node.
river reservoir. In this method, a feed-forward or multi-layer k1(.) and k1(.) are the functions for activation.
feed-forward network is used which is trained by a back- The most commonly used function for activation is a sig-
propagation algorithm [36]. Then the three layers of propaga- moidal logistic function that is represented as:
tion of the neural network that can be mathematically 1
expressed as. KðyÞ ¼
1 þ ey
!!
X c Xi
The ANN model is designed and the procedure to be fol-
Apl ¼ k1 a
Hd;l K2 w
Hb;d yps lowed is shown in Fig. 4. The information data is initially
d¼1 d¼1
explained and evaluated on the ANN model of input informa-
9292 A.I. Taloba

tion with the outcome is desired. The information data is cat- process of training and the testing is performed until the suit-
egorized on moreover the validation set or the training set. able result was attained, when the suitable result was attained,
Thus, the ANN model uses the training set only in the process then the model of ANN is considered to have been qualified
of learning to develop the model. Thus, the set of validation is and then ready to use as shown in Algorithm 1.
used for model testing on its ability of predictive, when the
ANN training is stopped. The ANN architecture was defined Algorithm 1. Water treatment process.
by the selection of hidden layers quantity that can be created
and the number of neurons for the individual hidden layer.
There are no static rules on defining the model of ANN and 1 Procedure TRAIN
its principles of the parameter, a ANNs large number is cre- 2 A Data set for Training o
p
ated with the various parameter and structures before deter- 3 B Labels for records in A
mining an acceptable model [37]. By the performance of 4 R The respective layer weights
5 n The layer number in the ANN, 1. . ..N
testing the periodic design of ANN on the set of tests and then
6 XðnÞ the error for all n,u,v
recording both the results of the test data and training results uv

set, thus the iteration number that introduces the best design
7 YðnÞ
uv o, for all n,u,v
8 For u = 1 to o
was attained [38]. They need to the ANN model reset and then
9 gn feedforward (a(u), z)
train the network increases to the iteration number. 10 hn g(N) – B(u)
The parameters of ANN are set when before the process of ðnÞ
11 YðnÞ
uv YðnÞ
uv þ gv :Yu
nþ1
training is started. This process of training includes the output
12 If v – o then
of computation from the input and weights. Thus, the algo- 13 XðnÞ
uv
1 ðnÞ
o Yuv þ kRuv
ðnÞ
rithm for backpropagation is used for ANN training that
14 Else
adjusts its weights that reduce the variation between ANN cur- 15 XðuvnÞ 1 ðnÞ
o Yuv
rent output and the output chosen [39]. At last, the calculation
16 End
of this process is conducted then to control, then the ANN
model has a study to resolve the method at one hand. Thus,
the process of evaluation is involved halting periodically the
In the ANN model, the process of accurate optimization is
attained by the different methods of the dosage of coagulant
and then the function of cost is minimum. The dosage of coag-
ulant was evaluated via sedimentation effluent turbidity, price
calculation, and ANN [40]. To obtain the suitable minimum
treatment of overall costs. The mean turbidity load is rapid fil-
ters influent that against the runtime filter, it is noted then a
load of mean decreases, then runtime filter increase is shown
in Fig. 5. Then equally varying the turbidity the load means
is attained from the load of turbidity that occurs within a
run filter [41].
The various relations between the runtime filter in hours
and then the turbidity means load in the filter that influent is
defined in below equation:
Runtime filter ¼ 957 ðTurbidity loadÞ0:72
where n as filters number, the price for filter backwash is eval-
uated as.
Priceusefilterbackwash
Pricefilterbackwash ¼ n 
957:ðTurbidityloadÞ0:72
The range of the parameter is monitored on the dosage of
coagulant till the price is matched on the minimum treatment
of prices. The pH level of water is 7.2 is observed during the
time range for part of treatment.

5. Result and discussion

The significant information data values are used to calculate

the better suitable values of the factor of weights and top sys-
tem model for the network. To progress the optimization to
reduce the cost for the water treatment process and the desali-
nation process, the cost for the chemicals, backwashing the fil-
ter, and discharge the waste is also considered. Price
minimization will be through the all-coagulation process, fil-
Fig. 4 Design for ANN model.
An artificial neural network mechanism 9293

Priceonebackwashfilter
price Backwash filter ¼ l Filter in operation:
runtime filter

From Fig. 6 shows the graph for the load of mean turbidity
that is influent of the rapid filters that against the runtime filter
and it can be noted that the load mean value is got decreases
and then the run time filter is got increases.
The dosage of the coagulant parameter range is got moni-
tored until the quantity of minimum matches the price of treat-
ment. The level of pH in water is measured as 7.2 during the
part of treatment in a range of time. Thus Fig. 7 shows the cal-
culated turbidity and the measured turbidity in the sedimenta-
tion effluent.
The performance of the estimated result is attained through
the help of the proposed concentration of coagulant is mea-
sured, then the optimization in dosage of coagulant by the pro-
cess of integrating with ANN, then the calculated and the
measured turbidity in the sedimentation effluent step after
the coagulant dosage of optimization ANN. Fig. 8. the dosage
coagulant of dosage by the price with the ANN optimization is
related to the coagulant is measured the concentration of coag-
ulant Fig. 9 shows the measured and observed turbidity is
expected in the sedimentation effluent after the coagulant opti-
mization by the ANN.
The ANN is used in a Multilayer network of perceptron
and it is trained with the algorithm of backpropagation. Then
the output of the ANN is the turbidity of the effluent sedimen-
tation process. The suitable value that was set up to 5000 and
20,000 value set of data is separated for the validation of the
model. Thus measured the accuracy is very high with the mean
Fig. 5 ANN Flow diagram for optimization. value of the absolute deviation of 1.3% with the training infor-
mation 1.2% and 1.4% for data validation. The wight process
for training or standardization can be taken as an important
tration, and sedimentation. The price function contains the time amount that depends on the network system model and
price for all coagulants (price coagulant), the price for energy the training algorithm.
(price energy), the price for sludge disposal (price sludge disposal),
and the price for backwash water filter (price backwash filter). 6. Conclusion

Price ¼ price coagulant þ price energy þ price sludge disposal

In this paper, the Artificial neural network (ANN) method is
þ price backwash filter utilized for cost optimization in the process of water treatment
and the desalination process was developed effectively by using
The coagulant price is x/h are then measured by adding the
this methodology. This paper offered some outcomes from
coagulant amount in Fe3+ in kg/h, then the specific price for
numerous practical ways utilized in water treatment facility
coagulant is.
operations, such as the utilization of sludge as a coagulant con-
Price coagulant ¼ Fe3þ : specpricefe 3þ tribution with aluminum sulfate in the water treatment plant.
The energy price (price energy) is measured by adding the
energy price for the stirrers and also for water pumping and
3.5
the air for the backwashing filter. The price was taken from
the digital meter then it will be installed along with the energy 3
source of stirrers and pump. The sludge disposal price is (price
2.5
sludge disposal x/h) is measured by adding the coagulant amount
(Fe3+ in kg/h), then the specific price for disposal (spec. price 2
3+
sludge disposal in x/kg Fe ). The sludge disposal price is paid at 1.5
a rate that is flat as it is dependent on adding the coagulant.
1
price sludge disposal ¼ Fe3þ : specprice sludge disposal
0.5
The backwash water filter price (price Backwash filter in x/h)
0
is measured by the one backwash filter price (one backwash 2007.5 2008 2008.5 2009 2009.5 2010 2010.5 2011 2011.5 2012 2012.5
filter price in x) is got divided by the runtime of filter and
y measured y calculated
then multiplied by the number of the filters in operations
(l Filter in operation). Fig. 6 Mean load in run times filter and rapid influent filter.
9294 A.I. Taloba

3.5 used as a suitable technique for analyzing the future work of

delivering drinking water using groundwater resources. More-
3
over, using the framework provided above, an artificial neural
2.5 network may be linked with numerical simulations that con-
2
tain physical rules to form a hybrid model for assessing and
forecasting intricate desalination procedures.
1.5

1 Declaration of Competing Interest

0.5 The authors declare that they have no known competing

0 financial interests or personal relationships that could have
2007 2008 2009 2010 2011 2012 2013 appeared to influence the work reported in this paper.
Turbidity measured Turbidity calculated
Acknowledgment
Fig. 7 Calculated and measured the turbidity.
The authors extend their appreciation to the Deanship of Sci-
entific Research at Jouf University for funding this work
Cost optimization with ANN through research grant No (DSR-2021-02-0105)
3.5
References
3

2.5 [1] G. Micale, L. Rizzuti, A. Cipollina, Eds., Seawater

Desalination: Conventional and Renewable Energy Processes,
2
Springer Berlin Heidelberg, Berlin, Heidelberg, 2009. doi:
1.5 https://doi.org/10.1007/978-3-642-01150-4.
[2] Q. Zhang, S.J. Stanley, Forecasting raw-water quality
1
parameters for the North Saskatchewan River by neural
0.5 network modeling, Water Res. 31 (9) (1997) 2340–2350.
[3] F. Qaderi, E. Babanezhad, Prediction of the groundwater
0
2007 2008 2009 2010 2011 2012 2013 remediation costs for drinking use based on quality of water
resource, using artificial neural network, J. Cleaner Prod. 161
Fig. 8 Cost optimization by using ANN. (Sep. 2017) 840–849, https://doi.org/10.1016/j.
jclepro.2017.05.187.
[4] M. Qasim, N.N. Darwish, S. Mhiyo, N.A. Darwish, N. Hilal,
3.5 The use of ultrasound to mitigate membrane fouling in
desalination and water treatment, Desalination 443 (2018)
3 143–164.
2.5 [5] T.D. Oyoh, Desalination in Water Treatment and
Sustainability, 2017.
2 [6] J. Jawad, A.H. Hawari, S. Zaidi, Modeling of forward osmosis
process using artificial neural networks (ANN) to predict the
1.5
permeate flux, Desalination 484 (2020) 114427.
1 [7] Y.-J. Choi, J.-S. Choi, H.-J. Oh, S. Lee, D.R. Yang, J.H. Kim,
Toward a combined system of forward osmosis and reverse
0.5
osmosis for seawater desalination, Desalination 247 (1-3) (2009)
0 239–246.
2007 2008 2009 2010 2011 2012 2013 [8] A.M. Aish, H.A. Zaqoot, S.M. Abdeljawad, Artificial neural
network approach for predicting reverse osmosis desalination
Turbidity measured Turbidity calculated
plants performance in the Gaza Strip, Desalination 367 (Jul.
2015) 240–247, https://doi.org/10.1016/j.desal.2015.04.008.
Fig. 9 Measured and observed the expected turbidity by ANN.
[9] M. Khayet, C. Cojocaru, Artificial neural network modeling and
optimization of desalination by air gap membrane distillation,
The water treatment process is used to produce the treated Sep. Purif. Technol. 86 (Feb. 2012) 171–182, https://doi.org/
10.1016/j.seppur.2011.11.001.
water for agriculture process, industrial process, and drinking.
[10] Q. Cai, B.C.Y. Lee, S.L. Ong, J. Hu, Application of a
This process must be done and that must against the price is Multiobjective Artificial Neural Network (ANN) in industrial
lowest that with the environmental impact is low. This tech- reverse osmosis concentrate treatment with a fluidized bed
nique usually includes water disinfection and water clarifica- fenton process: performance prediction and process
tion. This developed model can enhance the process for optimization, ACS ES T Water 1 (4) (2021) 847–858.
performance and the price is effective in water treatment using [11] W. Cao, Q. Liu, Y. Wang, I.M. Mujtaba, Modeling and
suitable information data. This research paper demonstrates simulation of VMD desalination process by ANN, Comput.
that the ANN can also be used for the optimization of water Chem. Eng. 84 (2016) 96–103.
treatment processes and thus the procedure can be automated.
Furthermore, depending on the model’s conclusions, it may be
An artificial neural network mechanism 9295

[12] Y.G. Lee et al., Artificial neural network model for optimizing [28] P. Gao, L. Zhang, K. Cheng, H. Zhang, A new approach to
operation of a seawater reverse osmosis desalination plant, p. 10, performance analysis of a seawater desalination system by an
2009. artificial neural network, Desalination 205 (1–3) (Feb. 2007)
[13] S. Zhang, Q. Tang, Y. Lin, Y. Tang, Fault detection of feed 147–155, https://doi.org/10.1016/j.desal.2006.03.549.
water treatment process using PCA-WD with parameter [29] K. Bourouni, T. Ben M’Barek, A. Al Taee, Design and
optimization, ISA Trans. 68 (May 2017) 313–326, https://doi. optimization of desalination reverse osmosis plants driven by
org/10.1016/j.isatra.2017.03.019. renewable energies using genetic algorithms, Renewable Energy
[14] Y. Choi, Y. Lee, K. Shin, Y. Park, S. Lee, Analysis of long-term 36 (3) (2011) 936–950.
performance of full-scale reverse osmosis desalination plant [30] M. Verhuelsdonk, T. Attenborough, O. Lex, T. Altmann,
using artificial neural network and tree model, Environ. Eng. Design and optimization of seawater reverse osmosis
Res. 25 (5) (Oct. 2019) 763–770, https://doi.org/10.4491/ desalination plants using special simulation software,
eer.2019.324. Desalination 250 (2) (Jan. 2010) 729–733, https://doi.org/
[15] M. Khayet, C. Cojocaru, Artificial neural network model for 10.1016/j.desal.2008.11.031.
desalination by sweeping gas membrane distillation, [31] M.G. Marcovecchio, S.F. Mussati, P.A. Aguirre, Nicolás J.,
Desalination 308 (Jan. 2013) 102–110, https://doi.org/10.1016/ Scenna, Optimization of hybrid desalination processes including
j.desal.2012.06.023. multi stage flash and reverse osmosis systems, Desalination 182
[16] D. Karimanzira, T. Rauschenbach, Deep learning based model (1-3) (2005) 111–122.
predictive control for a reverse osmosis desalination plant, [32] N. Gupta et al, Artificial neural network, Network Complex
JAMP 08 (12) (2020) 2713–2731, https://doi.org/ Syst. 3 (1) (2013) 24–28.
10.4236/jamp.2020.812201. [33] S. Al Aani, T. Bonny, S.W. Hasan, N. Hilal, ‘‘Can machine
[17] S.M. Kargar, R. Mehrad, Robust model predictive control for a language and artificial intelligence revolutionize process
small reverse osmosis desalination unit subject to uncertainty automation for water treatment and desalination?,
and actuator fault, Water Supply 20 (4) (Jun. 2020) 1229–1240, Desalination 458 (2019) 84–96, https://doiorg/10.1016/j.
https://doi.org/10.2166/ws.2020.043. desal.2019.02.005.
[18] F. Manenti et al., Operational optimization of reverse osmosis [34] D.-J. Choi, H. Park, A hybrid artificial neural network as a
plant using MPC, 2015. software sensor for optimal control of a wastewater treatment
[19] H.-G. Han, L.u. Zhang, H.-X. Liu, J.-F. Qiao, Multiobjective process, Water Res. 35 (16) (2001) 3959–3967.
design of fuzzy neural network controller for wastewater [35] M. Huggi, S. Mise, ANN model of Wastewater Treatment
treatment process, Appl. Soft Comput. 67 (2018) 467–478. Process, Ann Model of Wastewater Treatment Process, pp. 1–
[20] M.A.K. Al-Sofi, A.M. Hassan, O.A. Hamed, A.G.I. Dalvi, M. 10, 2019.
N.M. Kither, G.M. Mustafa, K. Bamardouf, Optimization of [36] P. Antwi, J. Li, P.O. Boadi, J. Meng, E.n. Shi, K. Deng, F.K.
hybridized seawater desalination process, Desalination 131 (1-3) Bondinuba, Estimation of biogas and methane yields in an
(2000) 147–156, https://doi.org/10.1016/S0011-9164(00)90015-8. UASB treating potato starch processing wastewater with
[21] C.S. Bandi, R. Uppaluri, A. Kumar, Global optimization of backpropagation artificial neural network, Bioresour. Technol.
MSF seawater desalination processes, Desalination 394 (Sep. 228 (2017) 106–115.
2016) 30–43, https://doi.org/10.1016/j.desal.2016.04.012. [37] A. Adgar, C. Cox, T. Böhme, Performance improvements at
[22] B. Wu, F. Hochstrasser, E. Akhondi, N. Ambauen, L. surface water treatment works using ANN-based automation
Tschirren, M. Burkhardt, A.G. Fane, W. Pronk, Optimization schemes, Chem. Eng. Res. Des. 78 (7) (2000) 1026–1039.
of gravity-driven membrane (GDM) filtration process for [38] R. Porrazzo, A. Cipollina, M. Galluzzo, G. Micale, A neural
seawater pretreatment, Water Res. 93 (2016) 133–140, https:// network-based optimizing control system for a seawater-
doi.org/10.1016/j.watres.2016.02.021. desalination solar-powered membrane distillation unit,
[23] L. Truttmann et al, Gravity-driven membrane (GDM) filtration Comput. Chem. Eng. 54 (Jul. 2013) 79–96, https://doi.org/
of algae-polluted surface water, J. Water Process Eng. 36 (2020) 10.1016/j.compchemeng.2013.03.015.
101257. [39] W. Wu, G.C. Dandy, H.R. Maier, Protocol for developing
[24] S.S. Madaeni, M. Shiri, A.R. Kurdian, Modeling, optimization, ANN models and its application to the assessment of the quality
and control of reverse osmosis water treatment in Kazeroon of the ANN model development process in drinking water
power plant using neural network, Chem. Eng. Commun. 202 quality modelling, Environ. Modell. Software 54 (2014) 108–
(1) (Jan. 2015) 6–14, https://doi.org/10.1080/ 127.
00986445.2013.828606. [40] T. Ogwueleka, F. Ogwueleka, Optimization of drinking water
[25] M. Farhaoui, M. Derraz, Review on Optimization of Drinking treatment processes using artificial neural network, Nigerian J.
Water Treatment Process, p. 10. Technol. 28 (1) (2009) 16–25.
[26] J.-Q. Jiang, The role of coagulation in water treatment, Curr. [41] Q.J. Zhang, A.A. Cudrak, R. Shariff, S.J. Stanley,
Opin. Chem. Eng. 8 (2015) 36–44. Implementing artificial neural network models for real-time
[27] J.J. Lee, J.H. Cha, R. Ben Aim, K.B. Han, C.W. Kim, Fiber water colour forecasting in a water treatment plant, J. Environ.
filter as an alternative to the process of flocculation– Eng. Sci. 3 (Supplement 1) (2004) S15–S23.
sedimentation for water treatment, Desalination 231 (1-3)
(2008) 323–331.