Global NEST Journal, Vol 26, No 2, 05481

Copyright© 2024 Global NEST

Printed in Greece. All rights reserved

Predictive modeling for solar desalination using artificial neural

network techniques: A review
Vairavel Madeshwaren1∗, Kathiresan Govindaraju2, Pragya Pandey3, Suresh Thangavel4 and Rajpradeesh Thangaraj5
1 Department of Mechanical Engineering, Annapoorna Engineering College, Salem, Tamil Nadu, India

2 Department of Mechanical Engineering, Excel Engineering College, Komarapalayam, Namakkal,Tamil Nadu, India

3 Department of Mathematics, SRM IST Ramapuram, Chennai, TamilNadu, India

4 Department of Mechanical Engineering, Kumaraguru College of Technology, Coimbatore, Tamil Nadu, India

5 Department of Mechanical Engineering, Kalasalingam Academy of Research and Education, Krishnankoil, TamilNadu, India

Received: 01/11/2023, Accepted: 01/01/2024, Available online: 13/01/2024

*to whom all correspondence should be addressed: e-mail: phdannauniv2020@gmail.com
https://doi.org/10.30955/gnj.0005481
these methods with the proper inputs and structures, it can be
Graphical abstract
deduced that the results of the solar desalination units can be
consistently and correctly projected. Additionally, several
suggestions are offered for future research in the relevant
areas of the study.
Keywords: Artificial neural network, solar still, ANFIS-based
model, genetic algorithm, support vector machine
1. Introduction
The globe is experiencing an energy crisis, and via scientific
research, researchers from all around the world are
working arduously to find solutions by focusing on
renewable energy sources. Energy is a crucial component
needed for human life to exist on Earth and is crucial to the
growth of human life. The ongoing rise in demand, the
expansion of the population, and the improvement of living
conditions are the causes of the energy problems. Because
fossil fuels are becoming scarcer and greenhouse gas
emissions are rising, it is getting harder to meet energy
needs with traditional resources. Utilizing multiple natural
or renewable resources is therefore necessary.
To meet one of the fundamental needs of any nation,
which is the rising energy demand. The amount of energy a
Abstract nation consumes determines its level of growth and
Due to the limitations of fossil fuels and the environmental advancement. The development of alternative energy
problems associated with their usage, renewable energy sources is essential for sure and steady advancement. The
sources have been exploited for desalination through the best choice for it is renewable energy. There are several
employment of various technologies and mediums. One of types of renewable energy, but the most significant source
the most useful renewable energy sources for solar of thermal energy is solar radiation.
desalination, both directly and indirectly, is solar energy. The Modern technology is getting increasingly concerned with
effectiveness of solar desalination is influenced by a variety of Thermal Energy Storage (TES). By storing thermal energy
parameters, making it challenging to forecast their during times when it is abundantly available and utilizing it
performance in particular circumstances. Artificial neural when and where it is needed, TES systems enhance energy
networks (ANNs), PSO, ANFIS, RO, and genetic algorithms management. TES is used in a variety of applications,
would all be suitable techniques for their modeling and including air conditioning, waste heat recovery, space and
output predictions in this context. In the current research, water heating, and more. It is therefore the most promising
multiple data-driven approaches are used in-depth to perform energy source. India is fortunate to have abundant solar
modeling of solar-based desalination facilities. By utilizing radiation available virtually all year round throughout its

Vairavel Madeshwaren, Kathiresan Govindaraju, Pragya Pandey, Suresh Thangavel and Rajpradeesh Thangaraj. (2024), Predictive
modeling for solar desalination using artificial neural network techniques: A review, Global NEST Journal, 26(2), 05481.
2 VAIRAVEL MADESHWAREN et al.

territory. Water heating is one of the current uses for solar procedures, designs, and operating factors on the
energy. Solar water stills are becoming more and more performance of a solar distillation unit.
popular because they are relatively cheap to make and
2. Artificial neural network
maintain. They serve as a substitute for or in addition to
electric or gas geysers. An area of Artificial Intelligence (AI) known as Artificial
Despite being cost-effective, non-polluting, and endless, Neural Networks is a subset of computational algorithms.
solar energy is time-dependent and has an intermittent Biological neural systems that build an input-output
nature. As a result, some type of TES is required, which mapping to learn from examples are the inspiration for
improves solar energy utilization and is just as crucial to the ANN models. It is a straightforward mathematical method
development of new energy sources. Systems using Phase made to handle several tasks. ANN is made up of
Change Materials (PCM) are more popular than other characteristics that result in ideal solutions. These qualities
options for energy storage because of their consistency in include mistake tolerance, parallel processing,
latent heat storage. However, the PCM for thermal energy generalization, and learning capacity. These characteristics
storage is still in the early stages of development. would enable the ANN to precisely and adaptably address
complicated issues.
Untreated water contains a variety of contaminants, such
as iron, arsenic, fluoride, and more, making it unfit for The basic transforming units (neurons) that make up an
human consumption. Over a billion people do not have ANN are numerous, interconnected, and layered. Input
access to clean drinking water, according to a United data and associated output values are required to train and
Nations report, and this number is expected to rise as the test a neural network. Five components make up an
world's population grows compared to its water supply. artificial neuron: weights, inputs, activation function,
Water covers about two thirds of the earth's surface. Still, output, and summation function. Figure 1 shows the
the accessible water is salted and so unfit for human structure of ANN.
consumption. According to reports, 97% of the water on
Earth's surface is salty. A significant amount of the 3% of
water on Earth that is not salty is found in the polar regions
as icebergs or as seawater. Less than 1% of this fresh water
is accessible to humans, which is insufficient to cover all
needs. Furthermore, 90% of urban sewage in developing
nations is dumped into waterways, creating a massive
amount of waste and converting the waterways into
sewers or sources of contaminated water. Currently, 884
million people do not have access to clean water supplies;
1 in 8 people do not have access to safe drinking water; and
every day, water-borne illnesses claim the lives of about
24,000 children under the age of five.
Some locations have far greater fluoride percentages than
the 1.5 parts per million (ppm) that the World Health
Organisation (WHO) recommends for drinking water. Many
methods have been devised to make use of the available Figure 1. ANN model
dirty water for the creation of safe and clean water. The W1, W2, and Wn are weights, whereas X1, X2, and Xn are
majority of advanced technology rely on active input values. The appropriate weight is multiplied by each
mechanisms that use enormous amounts of power input. The output of the neuron is derived by applying the
produced by burning fossil fuels. In the next two decades, activation function to the result of the summary. Learning
the global demand for fossil fuels is predicted to surpass normally takes place during a certain training period in an
the yearly production. A lack of petrol or oil can potentially artificial neural network. The network enters a production
spark violence and international economic and political phase where it generates results on its own after training.
issues. Furthermore, burning fossil fuels creates toxic A static network is a system that has separate learning and
emissions that have an impact on the local, regional, and production phases. Dynamic networks are those that can
global environments. These emissions include carbon continue to learn while they are being produced. However,
dioxide, nitrogen oxides, and aerosols.Therefore, it is the type of application and data format of a particular
challenging to produce enough drinkable water at a problem determine the best artificial neural network
reasonable cost in areas where grid power has not yet been topology to use.
reached.Many technologies have been developed to far, 2.1. Types of ann
primarily in underdeveloped nations, to use solar energy to
create distillate; however, the efficiency of these systems 2.1.1. Single-layer feed-forward (SFF) neural network
are found to be rather low. To help establish an effective When the input layer of node sources does not project into
model for higher yield and higher-quality distillate, an the output layer of neurons, the neural network is referred
attempt has been made in this work to analyse various to as a (SFF) or acyclic (NN). In a one-layer network, the
PREDICTIVE MODELING FOR SOLAR DESALINATION USING ARTIFICIAL NEURAL NETWORK 3

output layer of computer nodes is known as the single amount of error. The next sections go into great detail
layer. regarding the training and testing phase.
2.1.2. Multiple layers feed-forward neural network
The computation nodes in this kind of network are divided
into one or more hidden layers and are referred to as
hidden neurons. Hidden neurons' roles include extracting
higher-order statistics and interacting beneficially between
network output and external input. Neurons in the second
layer of the network receive their input signal from source
nodes in the input layer. The second layer's output signals
are fed into the third layer, and so forth. The overall
response to the pattern of activation is made up of the set
of output signals from the neurons in the output layer of
the neural network, which improves prediction accuracy.
2.1.3. Back Propagation (BP) Algorithm
The backpropagation (BP) learning algorithm is used in
multi-layer neural network architecture to train the input
and target pair patterns in a supervised way and divided
into two trips over the various network layers: Backward Figure 2. Feed forward back propagation model
pass and forward pass
3.1. Training Phase
The weights between neurons are adjusted layer by layer
when input data is transmitted through the system forward Training and neural network designs are carried out during
during the FP (forward pass). The network's real response the training phase. The subsequent substeps are
is then produced as a sequence of outputs. The networks' employed:
synaptic masses are fixed during the FP. An error correction • Acquiring the data.
rule adjusts the weights during the reverse pass. • Splitting the data (for example, 80% for training and
An error signal is produced by deducting the network's 20% for testing).
actual output from its predicted response. This incorrect • Normalization of data from origin to destination.
signal is then reverse-transmitted by the network. The
prediction error is decreased across several training cycles • Choose the different parameters.
until the network gets the required degree of accuracy. A • NN size.
full cycle of forward-backward passes and weight
• The transfer function type that will be applied to each
adjustments using each input-output pair in the data set is
layer.
referred to as an epoch, also known as an iteration. The
network consistently reaches the predetermined level of • The algorithm used for training.
accuracy. The idea of feed forward-back propagation is • Identifying the best-hidden layer and hidden neurons.
demonstrated in Figure 2.
• Use the training dataset to run the neural network
Where, through its learning process.
X1, X2, X3, ..., Xi are inputs • In the event of any of the following, training is
W1,4, W1,5, Wi,j, …, Wj,k are weightsO7, O8, O9, …, Ok are terminated:
outputs • Achieves the performance objective.
3. Design and development of ann model • The maximum number of repeats (epochs) has been
The creation of ANN models goes through three crucial achieved.
stages. Phases of Training, Testing, and Performance • The allotted time has been exceeded.
Evaluation sending a certain set of inputs through the
• The performance gradient is below the established
network and comparing the results with a specific set of
minimum value.
intended outputs is how training is carried out. The weights
are changed to generate a set of outputs that are closer to 3.2. Testing phase
the goal values if there is a discrepancy between the actual In the testing step, the trained ANN model is assessed
and target outputs. The old weight is multiplied by the against testing data to determine its suitability for
error correction value to create the network masses, and achieving the objective of accurate performance with little
the bias value is also changed. Up till the performance goal error.
is reached, this process is repeated. The created ANN
4. Performance evaluation
model is finalized against the testing data after the training
phase is complete to meet our requirements with the least The effectiveness of the ANN is assessed using the
following parameters to calculate the statistical error
4 VAIRAVEL MADESHWAREN et al.

between the predicted and goal: RMSE, R-value, complete this project, they built two different ANN models
Correlation Coefficient, and Regression Analysis - MAPE using the 13-5-1 and 14-7-2 neural models. The model's
4.1. Solar still using ann's analysis training phase employed the data that had been collected.
R2 was 0.9945 for the first model and 0.9825 and 0.9910
From the aforementioned literature, it can be seen that for the second model for results that were expected. For
some researchers have created mathematical models the two models, the greatest percentage variances were
while others have carried out thorough experimental 1.9% and 5.5%, respectively (Kalogirou, 2000).
experiments to assess the effectiveness of various solar
heating methods. Due to lengthy computer programming 4.5. Hottel-Vhillier (H-V) model for solar still
codes, the analytical approach takes a long time to produce ANN model to forecast how well a flat plate solar still would
a suitable result, especially for the solution of complex function. Three input variables—ambient temperature, air
equations. To obtain results and draw useful conclusions, intake temperature, and solar intensity—and one output
the experimental study also needs a significant amount of variable—air outlet temperature—were used to build the
time for experimentation and analysis. Contrarily, the use ANN model. The LM learning method was used to train the
of ANN delivers significant information patterns in a multi- model. In the hidden and output layers, respectively, the
dimensional information domain while also saving time; as tang and purlin transfer functions were employed. The
a result, this technology is becoming more and more heat network model's measured data from 17 days of data
common in the scientific and engineering disciplines. In and the Hottel-Vhillier (H-V) model's generated data were
recent years, several scholars have used ANN to predict the both used to train the model. Finally, they believed that
performance of various solar stills. Without the need for output temperatures from three different kinds of solar
intricate experiments or explicit equations, ANNs can collectors would be adequate. They found that there was
represent complex systems. ANNs can predict the desired an average variance of 0.9 C in the outlet temperature of
output of a system when sufficient experimental data is the solar collector (Farkas & Geczy, 2003).
presented. Numerous research have used ANN to simulate 4.6. Multilayer perceptron (MLP) and Radial Basis Function
and predict the thermal characteristics of solar energy (RBF)
collectors. A review of the literature on applying the ANN
technique to forecast the thermal performance of various To estimate the useful heat gained and thermal efficiency,
solar stills is given below: an ANN model was applied to two different types of hybrid
solar stills. They built two different kinds of models,
4.2. ANN+ learning algorithm including MLP and RBF. The best MLP model for thermal
ANN to analyze the ability of a parabolic collector solar efficiency and useable heat gain in tube-type collectors was
steam generator. 396 patterns were gathered from trials found to have 9 hidden nodes. Due to the lowest MSE and
on a solar steam generation setup. They built an 8-8-4 ANN in the instance of the RBF model with 8-91-2, 6 and 3
model using the parameters. Three layers were employed neurons in the hidden layer were found to be the optimum
in the hidden layer. 349 of the 396 data patterns were used model for the pipe-type collector in terms of usable heat
for training and 47 for testing, out of which 349 were used. gain and thermal efficiency. Due to having the lowest MSE
A learning technique was used to train the ANN model, and and an R2 value larger than 0.95, the MLP model is deemed
it correctly predicted results with a maximum range of 4.8% to be marginally superior to the RBF model (Falcao et al.
and an R2 value of 1.678 (Kalogirou et al. 1998). 2004).
p
4.3. Neural Model analysis using hidden layers
ANN was recommended and used to evaluate the
effectiveness of a home solar still. To perform an ANN
analysis, they gathered data. They then created a 9- 19-5
neural model with three hidden layer layers. Using the
collected data to train an ANN model, two output
parameters had expected values of 0.8967 and 0.8994. The
actual data collected was 7.1% and 9.7%, respectively, of
what was anticipated (Kalogirou et al. 1999).
ANN model with 8-28-3 neurons for thermo-siphon solar
water heating system thermal performance prediction.
They gathered 54 data sets for this project, of which 46
were used for training and the remaining 8 for testing. The
ANN model was trained using a learning technique, and it
was able to predict outcomes with maximum variations of
1MJ and 2.2oC for two output parameters (Kalogirou et al.
1999).
4.4. Performance of solar collector using ANN
The ANN approach to forecast the effectiveness of a
household forced circulation solar collector still. To Figure 3. Experimental setup with cylindrical collector
PREDICTIVE MODELING FOR SOLAR DESALINATION USING ARTIFICIAL NEURAL NETWORK 5

An ANN model was created to forecast the performance of Essentially, it consists of a glass cover that allows for energy
the solar still absorber. To do this, they created an transfer, an absorber plate that holds saline or brackish
experimental setting (Figure. 3) and gathered data for the water, and a metallic frame with a black base to support
research. 40 data points total were gathered, of which 26 the water and cover. Blackening the base's interior surface
were used for testing and 6 for training. When they created allows it to efficiently absorb solar radiation that strikes it.
an ANN model with five variables in the input layer and four In order to be purified, the brackish water is fed into the
parameters in the output layer, they discovered that 7 basin. At the bottom of the glass cover is a place to catch
neurons in the hidden layer were the ideal number. The LM the distillate. The two main classifications of solar
learning approach was used to train the 5-8-4 neural distillation systems, such as passive and active solar stills,
model, which was successful in accurately predicting can be generally categorised. Glass is utilised as a glazing
outcomes with low error and high R2 values (Cetiner, material in the majority of the literature because it can
2005). transmit over 90% of incident short wave radiation.
4.7. Solar collector analysis using SISO and MISO model However, because it acts as an opaque material for long
wave radiation, glass has a low transmittance to long wave
Two models: multiple input, single output (MISO) and heat radiation of wavelengths between 5 and 50 μm that is
single input, single output (SISO). In contrast to the MISO emitted by the absorber plate. There have also been
model, which used thermal heat loss coefficient as both reports of the usage of plastic sheets and films as glazing
the output and the input data, the SISO model used solar material. A transparent plastic cover in the shape of a
radiation as both the output and the input data (Lecoeuche hemispherical dome that has an absorptivity of 0.9 and
& Lalot, 2005). ANN tool to estimate flat plate solar transmissivity of 0.8 is utilised by many authors. Sometimes
collector performance characteristics (Kalogirou, 2006). a very thin (few microns) dielectric substance is put on the
4.8. Different angles of solar still with ANN analysis glass cover to reduce reflectance. Utilising dielectric
The ANN model to forecast flat plate solar collectors' material has been found to reduce reflectance by up to
thermal performance. They used an experimental setup 50%.
(Figure. 4) for this work and gathered data to build an ANN 4.9. Direct Expansion Solar Aided Heat Pump's (DXSAHP)
model. The experimental time, date, solar radiation performance
intensity, absorber surface temperature, tilt angle, ANN technique to forecast a direct expansion solar-aided
azimuth angle, and declination angle were seven of the heat pump's (DXSAHP) performance. In Calicut, India, they
input factors that were chosen. The output parameter used built their 1st experimental arrangement and gathered
was the collector's thermal efficiency. In the intermediate data under various climatic circumstances. They first
layer, 20 neurons were chosen from two buried layers. In created a 2-10-5 neural model with an FFBP network for
theneural model, the logistic sigmoid transfer function and investigation by the ANN technique. This research used a
the back-propagation learning technique were employed. total of 60 data sets, of which 10 were used for testing and
The 7-20-20-1 neural model was successful in predicting 50 were used for training (Mohanraj et al. 2008)
the flat plate solar collector's thermal efficiency. The
highest and minimum variances of the findings were 4.10. Feedforward neural network for solar still
discovered to be 2.558484 and 0.001969, respectively Neural network technology to estimate the solar air's
(Sozen et al. 2008). thermal efficiency. Three distinct types of absorber plates
were used in the double flow SAH experiments, as
illustrated in Figure 5. Using the experimentally
determined parameters, they created the FNN and ANN
models. In the ANN model, three distinct learning
algorithms, including the LM, SCG, and CGP, were utilized,
however, in the FNN model, just the LM technique was
applied. In the ANN and FNN models, respectively, the
tang and more wavelet activation functions were applied.
Due to its lowesterror and maximum R2 value among the
three algorithms, the LM with 6-4-2 was determined to be
the best model for ANN for a mass flow rate. The best
model was LM with 6-4-2 for 0.08 kg/s. The FNN 6-5-3
model, which had the lowest error and best R2 value, was
also discovered to be a better FNN model for prediction
(Esen et al. 2009).
4.11. Levenberg-Marquardt (LM) algorithm
Direct growth ANN was used to forecast the energy analysis
of solar desalination techniques. For this investigation,
Figure 4 Experimental setup of liquid flat plate collector they set up a laboratory environment and collected data in
A solar still is a basic apparatus that operates using the Calicut, India, under various meteorological circumstances.
principles of water evaporation and condensation. They developed a pair of models for energy consumption
6 VAIRAVEL MADESHWAREN et al.

and efficiency using 3-13-6 neurons. The LM learning 4.13. Backpropagation (BP)
algorithm was used to train the neural model, and it To forecast the performance of solar collectors, an ANN
correctly predicted outcomes with lower COV and RMSE model was created. In Beijing, they carried out experiments
values and greater coefficients of correlation (Mohanraj et and gathered data. The ANN model has been created. Two
al. 2009). variables, such as heating capacity and efficiency, were
used in the output layers. The ANN structure was modeled
using the BP learning approach and the logistic sigmoid
transfer function, demonstrating that the model
performed as predicted (Xie et al. 2009).
5. ANN methodology for temperature prediction
ANN methodology for temperature prediction of water
during solar purifiers. They tested various training months
over nine days. Two neurons in the input layer of the ANN
model, such as solar radiation and outside air temperature,
were constructed. The temperature of the water was
employed in the output layer. Five alternative training
methods were used to train the ANN model. They
discovered that the best model for predicting food
temperature was the 3-5-2 neural network with TRAINRP
training techniques (Tripathy & Kumar, 2009).
Soft computing techniques to forecast solar collector
Figure 5. Type I, Type II, and Type III (absorber plates) performance using PCM materials (Figure. 7). ANFIS, SVM,
and ANN were three different kinds of soft computing tools
4.12. Stochastic Gradient Descent (SGD)
that were utilized. One of the important tools utilized to
Still images and the ANN approach to predict how well the forecast solar collector performance was ANN. Three layers
solar desalination process would work. Using the SGD of the hidden layer were used to build an ANN network
method, they ran tests and gathered 60 data sets. Fifty using the 5-7-1 neural model. They used the LM learning
were used for training, while ten were used for testing. algorithm to predict outcomes, and they achieved good R2
ANN model built using 3-12-5 neuron FFBP network. SGD values between 0.832 and 0.899(Varol et al. 2011).
was one of three different training algorithms that were
5.1. Open-Cycle solar regenerator using ANN
used. In Figure 6, They found that the LM algorithm for
learning was the best one using statistical error analysis. To forecast the ability of an open-cycle solar regenerator,
They forecasted outcomes with minimum RMSE and COV an ANN model of 2-10-1 neurons was utilized. (Figure 7)
values and maximum correlation coefficients of 0.9999 They made predictions with the least amount of error and
(Moahanraj et al. 2009). the highest correlation coefficient values (Al et al. 2011).

Figure 7. Solar absorber experimental arrangement using PCM

components (Kicsiny& Richárd, 2014).
5.2. Tansig transfer function of ANN to analyze solar still
performance:
ANN tool to analyze the performance of solar still. In
August and September, they carried out experiments in
Turkey from 10:00 a.m. to 5:00 p.m. for 5 days. They
employed flat and zigzag absorber plates, two different
varieties. The experiments produced a total of 80 data sets.
The ANN model was created using MLP 8-20-1 and eight
Figure 6. Photo of the arrangement (Alizami et al. 2020) showing
factors, including the type of absorber model, the time of
the solar still in (a) and the glazed absorber in (b).
the tests, the exit and inlet air temperatures, the
PREDICTIVE MODELING FOR SOLAR DESALINATION USING ARTIFICIAL NEURAL NETWORK 7

temperature of stored water, the surrounding temperature carried out experiments and gathered data. A total of 567
and absorber surface temperatures, and the solar intensity. data patterns were gathered, of which 80% were utilized to
The collector's efficiency was chosen as the output train the NN model and 20% to test it. The mean storage
criterion. Algorithms for backpropagation LM learning tank temperature, ambient temperature, solar radiation,
were used to structure the ANN model. It was employed in tilt angle, and thermal efficiency were the four input
the hidden layer of this model. Finally, they estimated the factors that made up the ANN model's structure. ANN
solar air collector's thermal efficiency and compared it to model built using FFBP techniques. To identify the
actual testing results. The neural model performed appropriate number of neurons for the buried layer, they
satisfactorily, as indicated by the values of SSE, MAE, MRE, examined models with 3 to 13 neurons. They found that
R2, and RMSE were determined (Caner et al. 2011). the LM with 13 neurons was the best ANN model due to its
5.3. Neutrophil-Lymphocyte Ratio (NLR) and Multiple low error. Finally, the neural model's prediction of the
Linear Regression (MLR)Models collector's thermal efficiency with a 4-12-1 accuracy was
correct. Additionally, they calculated the thermal efficiency
FFNN with the help of NLR and MLR to forecast the rate at using the ANFIS approach. They claimed that because the
which seedy grapes would dry. They performed tests using ANN approach has lower RMSE and COV values and higher
the setup (Figure 8), which was built using two different R2 values (0.811914), it is superior to the ANFIS method
kinds of SAC: PCM-based collector and extended surface. (Dikmen et al. 2013).
Data was gathered for the ANN model. The ideal ANN
model for predicting outcomes was created using 3-10-1 The ANN method to calculate the solar air collector's
neurons. When the FNN model's performance was thermal efficiency. He experimented with corrugated and
compared to that of the MLR and NLR models, it was trapeze-shaped absorber plates, two different varieties.
discovered that the FNN model performed better (Cakmak Experiments were carried out in Turkey in October
& Yildiz, 2011). between 9:00 and 17:00. 66 samples of data total were
obtained from the studies. ANN model was built using the
BP algorithm and the FF structure. The optimum ANN
model was chosen using 3 to 9 neurons in the hidden layer
based on a trial-and-error process. Researchers find that
the ideal topology for SAC analysis is LM-3. R2 values of
1.2345 and 1.5672 for the first type and second type
respectively, offer good model results (Benli, 2013).
The ANN method to forecast the solar PV/T system's ideal
thermal and electricity output. They carried out trials with
the PV/T setup and gathered information for the ANN
model. With the use of 2-5-1 neurons, an ANN model that
predicted data more accurately was created (Ammar et al.
2013).
Figure 8. Experimental set-up 5.6. MATLAB software for solar thermal still analysis
5.4. Nonlinear Autoregressive with eXogenous input (NARx)
Flat plate solar still heat transfer analysis was predicted
The ANN approach to forecasting the performance of using an ANN model with 6-30-6 neurons. They used
cutting-edge solar collector modeling. They built the 5-5-1 MATLAB software to compute 2509 data utilising an
and 5-4-1 NARX neural models, respectively, and optimisation strategy for this project. The computed data
experimented with performance-based assessments by EN were divided into three groups: validation (20%), testing
12975-2, a European Standard. The results demonstrated (20%), and training (80%). The neural model used a hidden
that the ANN model performed satisfactorily when layer of 20 neurons and a durable back propagation
combined with cutting-edge modeling (Fisher et al. 2012). learning approach (Hamdan et al. 2014).
To analyze the ability of a hybrid Photovoltaic double-pass The ANN approach to forecast how well massive solar
air heating collector, an ANN model was built. Data was systems will function. Over ANN modeling, experimental
obtained, and it was utilized for testing. 2000 data sets data were gathered over 226 days. The ANN model, which
altogether were gathered from Indian states. The ANN includes three additional levels and layers overall, was
model used FFBP methods to organize its performance constructed with five neurons in the hidden layer. An ANN
prediction. 18 neurons in a hidden layer LM were selected. model was developed using experimental data and a
In the end, the results were correctly predicted by the 4- learning algorithm, and it accurately predicted outcomes
15-4 optimal model. The RMSE values for thermal, (Kalogirou et al. 2014)
electrical, overall, and total thermal energy were found to
5.7. Solar still thermal energy systems (SSTES) with ANN
vary from 1.10 to 2.89%, 1.12 to 2.90%, 1.23 to 3.21%, and
1.209 to 2.90%, respectively (Kamathania & Tiwari, 2012). For estimating the performance of solar thermal still
performance with thermal system, an ANN model has been
5.5. ANN with ANFIS
implemented. From March 2011 to December 2012, they
ANN and ANFIS techniques to forecast evacuated tube carried out trials with the STES setup and gathered data
solar collector performance. For training and testing, they regarding the local climate in Ottawa. ANN models
8 VAIRAVEL MADESHWAREN et al.

incorporating FFBP networks were created utilizing two interior space and provide fresh water. The model's inputs
distinct learning methods, such as LM and SCG. It was were the seawater greenhouse's breadth and length, the
discovered that the 10-20-8 neural model and LM learning height of the front evaporator, and the transparency of the
algorithm were the best options (Yaïci & Entchev, 2014). roof. The model's output was the system's water yield. We
To use experimental data to forecast how well a ground looked at their various concealed one- and two-layer
source heat pump system will function, ANFIA and ANN architectures. They discovered that using one hidden layer
models were developed. Twelve inputs and one outcome with 9 neurons produced the maximum degree of accuracy,
parameter were used to set up the ANN and ANFIS models. as measured by R2, of 0.998. In addition to the model's
For the ANFIA and ANN models, we used backpropagation architecture, the functions used and optimization
and hybrid learning approaches, respectively. They techniques for hidden layers (Faegh et al. 2021) must also
discovered that the ANFIS performance prediction is be taken into account; however, it must be remembered
superior to the ANN model (Esen et al. 2015). that a rise in the number of layers that are hidden may
result in overfitting.
5.8. Multiple-layer perceptron neural network (MLPNN)
model 6.1. ANFIS-based models
ANN approach to forecast a PV-T evaporator's One of the most crucial elements affecting how accurately
performance in a solar still. For ANN modeling, they data-driven methods predict the results of solar stills is the
acquired experimental data. An ANN model with 3–18–5 applied method and algorithm (Rezk (2022); Mashaly
neurons was created. The Multiple-layer perceptron neural (2017); Nassef (2020); Olabi(2023). ANN-ANFIS-based
network model has the highest R2 and the lowest models, multilinear regression, and random forest (RF).
estimated variance and COV values when predicting the When compared to other methods, they discovered that
outcomes (Gunasekar et al. 2015). employing RF produced the most accurate predictions
(Mashaly et al. 2018). Another study evaluated the
Three different types of ANN models to forecast solar
effectiveness of classical ANN and SVM with the Harris
radiation, including the MLP, GRNN, and RBF models and
Hawk optimizer in forecasting the efficiency of an active
compared the results to data predicted by the Improved
solar still. Their models took into account inputs such as
Bristow-Campbell (IBC) model (Wang et al. 2016).
time, ambient temperature, solar irradiation, wind speed,
The ANN technique has been successfully employed for and vapor velocity (Zayed et al. 2021). They discovered that
performance prediction of solar still desalination since it ANN performed better than models based on ANFIS and
operates more quickly and forecasts the data with less may be improved even more by applying the optimizer. R-
time, according to a review of the aforementioned squared values for the ANFIS and ANN-based models in
literature. As a result, it can be concluded that the ANN their study were 0.9894 and 0.9983. Better adjusting of the
method is suitable for predicting the capability of solar parameters impacting the ability of the modeling
desalination techniques. technique can be attributed to increased accuracy of the
Table 1 provides a selection of research papers that apply simulations via the coupling optimizer. In a different study,
artificial neural networks to forecast solar still desalination the effectiveness of ANN, Multiple Regression (MR), and
performance ANFIS was examined while predicting the efficiency of an
inclined passive solar still. Relative humidity, solar
6. Data-driven techniques in solar desalination radiation, TDS, and feed flow rates of brine and feed were
Heat, electrical methods, and pressure are the three basic all included as inputs in the suggested models. Another
ways utilized for desalination. The oldest method for important consideration is the function that is used in the
boiling highly salinized water and collecting the steam that design of data-driven procedures.
results is thermal distillation. Now that the steam has been
collected and condensed, it may be used to make fresh
water. The salt and water are separated using an electrical
current in the electrical method. Pressure is the driving
force behind the selective porous membrane used in the
RO kind of desalination, which allows water to pass through
while leaving the salt behind (Aljuyi et al. 2021). The
thermal and RO kinds of desalination account for almost all
of the market. Even though RO systems make up almost all
of the installed capacity for desalination systems, thermal
desalination has some advantages. For instance, using
plant waste heat for the desalination thermal units might
increase the system's overall efficiency. Studies on the
utilization of data-driven techniques in SD processes have
largely concentrated on thermal kinds (Soleimanzade et al.
2022) For example, (Behnam et al. 2022) used ANN to
simulate the results of a humidification-dehumidification- Figure 9. A solar desalination system featuring a still, PV, and a
type SD that was used to humidify the greenhouse's collector.
PREDICTIVE MODELING FOR SOLAR DESALINATION USING ARTIFICIAL NEURAL NETWORK 9

As an illustration, (Bahiraei et al. 2021) tested various approaches for predicting the hourly water output and
membership functions in ANFIS-based models, such as the water temperature, respectively (Abubaker et al. 2021).
Pi curve, trapezoid, triangle, and differential between two Although they took a novel method for dynamic modeling
sigmoidal functions, to propose a model with the maximum of SD, their model's accuracy was confined and may be
exactness. (Figure 9)Regression's correlation coefficient for increased by taking into consideration additional elements
training data sets employing these approaches was like wind speed and feed temperature.
approximately 0.999. The most appropriate function in a
8. Genetic algorithm
network's structure for modeling can depend on the
physics of the issue, which can be discovered by putting The performance of solar stills can be enhanced by using
several functions to the test (Perendecie et al. 2008; Esmaili nanofluids. These solar stills can be accurately evaluated
et al. 2021; Perez et al. 2010; Kharab et al. 2014). using innovative techniques. Different data-driven
modeling techniques, such as ANN-GA, to simulate the
7. Linear regression analysis model
capabilities of slope solar still using carbon black nanofluid
It is crucial to take into account all of the useful at a concentration of 1.5% wt (Garud et al. 2021). The
components as inputs when modeling the system using suggested model's inputs included solar radiation, ambient
data-driven techniques. To achieve greater accuracy or air temperature, vapor temperature, wind speed, glass
increase comprehensiveness, several models have outlet and intake temperatures, and basin temperature. To
included new inputs. In addition to the variables used in the fine-tune the methods and provide the results with the
bulk of the research, (Abuella et al. 2015) employed best degree of accuracy, the models were combined using
broader variables including the number of hours in a day, the Bayesian optimization technique. They discovered that
day, and month numbers, cloud cover, and the differential while all of the suggested models could forecast the
in temperature between the inner and outer surfaces of system's performance with a fair amount of accuracy, using
glass as an example. They used a regression, cascaded RF produced the best results (Saadaoui eti al, 2021). The
forward ANN, and linear model with various numbers of efficiency of a nanofluidic solar still combined with a
neurons in their research. They discovered that the ANN thermoelectric module, for instance, was modeled using
model provided a more accurate forecast of the system's ANN in conjunction with the Imperialist Competition
production. Although this model is more thorough than the Method(ICM), genetic algorithm (GA), and other
ones stated earlier, it may still be improved by taking into techniques by (Jervase et al. 2001).
account additional elements like the system's
specifications, such as the sizes of various sections and the
material's qualities that determine how well the systems
function (Ibrahim et al. 2012; Alizamir et al. 2020;
Skumanich et al. 1975; Ramedani et al. 2014).
SD can be combined with other elements to increase
output. Forecasting the performance of these systems can
be done using data-driven methods (Kicsiny & Richard,
2014). The panel was used to power the still in the tank that
was used to warm the salt water before it was introduced
to the solar still (Wahbah et al. 2019). Saltwater was heated
further in the collector before going into the still. Figure 10
displays the system's schematic. They discovered that the
maximum model accuracy was achieved while utilizing 24
neurons, with an R2 of 0.987 after testing several network
designs and hidden layer neuron counts. Hybrid technology
would require more inputs from the systems, making the
modeling process more challenging (Ruivo et al. 2022;
Bocco et al. 2012; Daut et al. 2011; Mahesh et al. 2022;
Devi et al. 2011).
It is possible to model the dynamic efficiency of solar
desalination systems using data-driven methodologies. A
solar still with an improved design's water temperature and
hourly production of water were estimated using a variety
of ANNs in a study conducted by (Chiteka et al. 2020),
comprising feedforward (FF), backpropagation (BP), linear Figure 10. Genetic algorithm for solar desalination PSO-ANFIS
SVR, Support Vector Regression (SVR), RBF, and RF. Their AND PSO-ANN
models' inputs included wind speed, outside temperature,
solar energy received, and basin water depth. When Coupling the aforementioned optimization approaches
compared to the predicted data and the corresponding real greatly enhanced the model's accuracy, with ICM having a
values, FF and RBF were shown to be the most successful greater impact on accuracy improvement. The modeling
10 VAIRAVEL MADESHWAREN et al.

methodology has an impact on how accurately nanofluidic (Khanna et al. 2015), the effectiveness of MLP ANN and
solar desalination values are projected to be. PSO-ANFIS MLR in foretelling the instantaneous thermal effectiveness
and PSO-ANN, for instance, were employed by (Khare et al. of a solar still was evaluated. In comparison to MLR, they
2013) to predict the performance of a solar still using Cu2O discovered that utilizing MLP ANN produced a model with
nanoparticles. The models' inputs were comparable to greater exactness. Due to its more sophisticated structure
those from earlier work, and the result of the created and ability to model complex systems more effectively,
model was system efficiency. They discovered that MLP ANN has a higher level of accuracy.
coupling the optimization methods increased exactness in
both types of models, but that PSO-ANFIS provided the
highest level of modeling accuracy. In a different study
Table 1. Artificial Neural Network technique in the solar desalination process.

Citation Neural Model Type of Network Algorithms The Work's Purpose

[Layek et al. approximation of solar collectors'daily

SISO, MISO MLPNN BP
2099] performance
estimating solar water stillperformance
[Saini, 2008] 4-7-3 MLPNN LM
with a collector.

[Karmare &
ANN: 8-9-1, 9-3-1, 9-6-1 To forecast the hybrid solarstill's
Tikekar, RBFNN, MLPNN BP, RBF
RBF:9-84-1 performance
2007]

[Jaurket et to evaluate the flat platesolar still's

3-7-1 MLPNN TRAINLM
al. 2006] performance.
[Bhagoria et Forced circulation solar water
13-5-1, 14-7-2 MLPNN BP
al, 2002] desalination performance estimation.

[Verma &
System for solar water desalination using
Prasad, 7-24-2 MLPNN BP
thermosiphon performanceanalysis
2000]

[Gupta et al. 8-18-2 (3 hiddenlayers to forecast how well a domestic solar still
MLPNN BP
1993] with 18 neurons) willfunction.

[Prasad & 8-8-4 (3 hiddenlayers with

MLPNN BP to calculate the solar still'sperformance.
Daini, 1988] 8 neurons)

[Kalogirou, estimating the temperature of the water

2-4-1 MLPNN SCG,CGP, BFG,LM, RP
2000] during the solar desalination procedure

[Prasad &
5-10-10-2 MLPNN BP To measure the effectiveness of solar still
Sahu, 2017)
[Sahu & To determine how well direct expansion
Prasad, 2-10-4 MLPNN LM,SCG,CGP solar-driven desalination techniques
2017] perform

[Prasad& Energy study of direct expansion solar-

2-12-5 MLPNN LM
Sahu, 2016] drivendesalination techniques prediction

[Sahu &
ANN: LM,SCG, CGPWNN: To forecast the solar still's thermal
Prasad, 6-4-2, 6-5-2 WNN, ANN
LM efficiency
2016]
[Behura et calculating a direct expansion solar still's
1-9-4 MLPNN LM
al. 2016] performance
[Behura et To determine how well flat plate solar
7-20-20-1 MLPNN BP
al. 2016] collectors function thermally
[Sing &
To forecast the flat plate solar still's
Siddhartha, Six differentmodel MLPNN BP
performance metrics
2014]

[Arcaklioglu To gauge the effectiveness of the solar

2-5-1 FINN LM
et al. 32004] PV/T system
PREDICTIVE MODELING FOR SOLAR DESALINATION USING ARTIFICIAL NEURAL NETWORK 11

[Pacheco et To calculate the thermal efficiency of two

8-3-1 MLPNN LM
al. 2001] different solarstill types
[Jani et al. To forecast how well an evacuated tube
4-12-1 ANFIS, FFNN LM
2017] solar still willoperate

[Zahraee et Calculate the hybrid PV/T double pass air

4-15-4 MLPNN LM
al. 2016] collector'sperformance
[Yadav & comparison of neural network and
Chandel, 5-5-1, 5-4-1 NARX LM cutting-edge solar stillmodeling
2014] techniques

[Karabacak& To calculate the rate of solar still

3-10-1 FINN LM
Cetin, 2014] performance inseawater.
[Uma
Maheswari &
8-20-1 MLPNN LM Solar still performance analysis
Meenakshi,
2018]
[Mellit et al, estimating open-cycle solar still
2-10-1 MLPNN BP
2009] performance.
[Kalogirou, determine the solar collector's
5-7-1 MLPNN LM
2001] performance based onPCM.
To forecast a PV-T evaporator's
[Yilmaz &
4-15-4 MLPNN LM performancefor solar-aided desalination
Atik, 2007]
methods
[Ertunc & To forecast the effectiveness of a ground-
13-15-1 MLPNN,ANN BP, Hybrid
Hosoz, 2006] basedsolar desalination system

[Hosoz & To gauge the effectiveness of the solar

10-20-8 FFBP LM,SCG
Ertunc, 2006] thermalenergy system

[Yigit & assessment of huge solar still kinds'

3-5-5-5-2 MLPNN BP
Ertunc, 2006] performance

[abbassi
To forecast the flat plate solar still's heat
&Bahar, 5-20-5 NARX Rprop
transferanalysis
2005]

employed at their optimum values, it is discovered that

9. RO-ANN
applying optimization algorithms and connecting them
The outcomes of the ANN-based model can be used to with intelligent approaches improves accuracy. The
design the ideal conditions for the efficiency of desalination examined inputs have an impact on the exactness in
systems (Ghermandi et al. 2009). As an example, in a study addition to the previously specified elements. Models will
done by (Ramanzanian et al. 2023), an ANN-based be more accurate if more significant inputs are taken into
optimizing control system was used to regulate a solar- account (Manolakos et al. 2008). The additional elements
powered transmembrane distillation module. After that, a that might contribute to the model's discrepancies can be
control system was put in place to maximize the system's considered to be the data noise that is inescapable in
distillate production. The suggested technology made it experimental data utilized for modeling.
possible to adjust the feed flow rate to achieve continuous 10. Dynamic performance of solar still using ANN
optimum distillate output at the ideal levels. Another
A cutting-edge technique for boosting the volume of
example is the way (Boesch & William, 1982) uses ANN to
desalinated water in solar desalination plants. Solar panels
forecast the weather and optimize a hybrid RO desalination
and a cylindrical parabolic collector (CPC) were employed
system that is powered by solar and wind energy. The
to achieve this goal of increasing the temperature of the
system's ideal design was created by using the network's
basin water. Investigations were also conducted into the
outputs and doing optimization.
effects of various basin solar still components on
The study's conclusions can be summed up by saying that freshwater mass. Of the several elements that make up a
the methods used, optimization algorithms, etc., have an basin, the aluminium basin has been linked to the greatest
impact on the models' accuracy. Due to their more amount of water desalination. In addition, the impacts of
sophisticated structures, which allow them to more various water depths and colours on the basin were
accurately mimic complex systems, intelligent approaches, investigated. At a 5-mm water depth, the maximum
like ANNs, are typically chosen in terms of accuracy. freshwater content in the black aluminium basin was 2.97
Additionally, as the parameters impacting exactness are kg/day (Bagheri et al. 2021).
12 VAIRAVEL MADESHWAREN et al.

These days, artificial intelligence is a major technological either be disposed of or transformed into a commodity that
advancement that can benefit business and research is valued. Products that have been mineralized may be
across a wide range of disciplines, including the usefully used in amounts that may eventually lessen the
desalination of water using solar thermal energy. consequences of global warming. There includes discussion
Therefore, the primary goal of the research is to forecast of typical rock kinds and how they react with CO2 gas.
the efficiency of water desalination, which is dependent on
12. Thermal conductivity in solar still
solar thermal energy, by using an artificial intelligence
regression model. Before implementing the system, the The thermal conductivity of solar still using ANN and a
contribution aids researchers and manufacturers in combination of water-based Al2O3 and CuO nanofluids
assessing the productivity of the desalination system's were assessed at different temperatures and volume
design in a beta setting (Salem et al. 2022). concentrations. Ethylene glycol and water are combined in
The research team's improved design for a solar still a base fluid mixture of 1:1 by weight. In addition, a
desalination system is taken into consideration here, and temperature range of 15 to 50 °C was obtained, and the
ANN models are developed for it using the experimental concentration was found to be 0.8% . In comparison to
data collected over the course of a year. One of the most Al2O3 and CuO nanofluids, the base fluids exhibit low
widely used machine learning techniques is the thermal conductivity, according to the test results that
development and comparison of various artificial neural were obtained. When compared to Al2O3 nanofluids, the
network (ANN) models in order to determine which CuO nanofluids exhibit higher thermal conductivity at the
performs best in predicting the two most important system same volume and temperature concentration.
performance parameters: the hourly produced distillate Nanoparticles suspended in any liquid provide energy to
and the water temperature in the basin (Sohani et al. nanofluids. The parameters of base fluid heat transmission
2021). were altered by the recent addition of nanoparticles. In this
experiment, the efficiency of distillation can enhance by up
To improve the efficiency of solar energy-powered solar to 29% by combining violet dye with water. Studies were
still units by utilising solar panels and cylindrical parabolic conducted on the novel nanofluid additives for solar still
collectors. Salinized water is heated by thermal performance. It was calculated how much the 1000 solar
components outside the solar still unit using a 300 W solar still performance units would weigh overall using ANN.
panel. Since solar panels are cooled during the hottest Based on the experiment, it was determined how
parts of the day, lowering their temperature may boost the introducing a carbon nanotube to the basin water affected
efficiency of the panels first and the solar still unit second. the hoover still's yield.
2.132 kg of freshwater per day was the largest amount
used in the experiment. ANNs were used to model the 13. Conclusion
experiments. Neural network modelling and experimental Applications of data-driven methodologies for modeling
data show a strong association, as indicated by the results solar desalination systems are givenin the paper. These
of neural network simulations (Bagheri et al. 2020). have been developed using a variety of inputs, such as
According to the performance evaluation criteria, the ANN ANN, ANFIS, PSO, RO, BP, GA, SVM, and others. The
model outperformed the MVR and SWR models. Compared following are the key conclusions of this review article:
to the MVR and SWR models, the ANN model's mean 1. When compared to correlation, intelligent approaches
coefficient of determination was around 13% and 14% can more precisely model solar desalination systems.
more accurate, respectively. Furthermore, the MVR and
SWR models' mean root mean square error values, at 2. By utilizing intelligent approaches, several parameters,
6.534% and 6.589%, respectively, were nearly twice as high including output, energy, and energy efficiency, may
as the ANN model's mean values. While the findings from be modeled.
the SWR and MVR models were comparable, the MVR 3. The applied approach and algorithm as well as the
model produced superior outcomes (Mashaly & Alaza et al. taken into account inputs are some of the factors that
2016). affect how accurate the proposed models are.
11. Minearlization process 4. Using optimization techniques in conjunction with the
models will increase accuracy because the
The natural process of mineralization is how the
hyperparameters will be set to their ideal values.
environment stores CO2, but it takes time. One of the main
causes of rock chemical weathering is the hydrolysis of CO2 5. The type of optimization algorithm has an impact on
in damp air or water. According to the geological record, the models' accuracy in addition to the modeling
weathering happens at a rate that can significantly lower approach used.
atmospheric CO2 levels when tectonic action exposes huge 6. Among the most crucial elements that must be
rock masses to the atmosphere. Even though natural employed as inputs are operating circumstances.
weathering may remove around 30 Gt of CO2 from the
7. The models' outputs, which were obtained via clever
atmosphere every century, this process might potentially
techniques, can be used to improvethe systems.
be accelerated and turned into an industrial process.
Anthropogenic CO2 can thus react with a reactive substrate 8. The majority of research has used water productivity
to create carbonate salts, a mineralized product that can as the model's output, but it would be advantageous
PREDICTIVE MODELING FOR SOLAR DESALINATION USING ARTIFICIAL NEURAL NETWORK 13

to take other technical factors like energy and the PSO-Particle Swarm Optimisation
system's efficiency into account.
Competing interests
9. It would be beneficial to take into account economic
and environmental elements as model outputs in The authors declare that they have no competing interests.
addition to technical criteria. Authors’ contribution
10. It is advised to compare the amount of time and Author 1 supports to comparison of existing and past
calculations needed for the training process of research literature. This helps to highlight our review
intelligent models using various algorithms and manuscript from others as an enhanced one.
methodologies.
Author 2 assists in writing content related to Solar still using
11. Models can be more accurate by using hybrid ANN analysis in our review paper. Author 3 helps and finds
optimization algorithms, which have a better capacity the performance of Solar still using ANN.
to find optimal solutions.
Author 4 and 5 supports and develops the studies of Soalr
The material for the basin has been chosen creatively, and still thermal energy system with ANN.
the distinctions between the materials have been
investigated. On several experiment days, heat Acknowledgment
characteristics are shown. If the right materials are I offer up our fervent prayers to the omnipotent God. I want to
employed, the temperature of the basin can be elevated, express my sincere gratitude to my co-workers for supporting me
which will increase the rate of desalination and the through all of our challenges and victories to get this task done. I
temperature of the water. During the process, this novel want to express my gratitude for our family's love and support, as
approach raises the temperature of the basin water by well as for their encouragement. Finally, I would like to extend our
utilising solar panels and an artificial neural network (ANN) sincere gratitude to everyone who has assisted us in writing this
in solar desalination facilities. There is typically good article.
agreement between the ANN-based prediction and the
References
experimental results when a variable number of neurons is
predicted based on the ANN inputs. ANNs have been used Abbassi, A., Bahar, L. (2005). Application of neural network for the
as one of the most dependable techniques for prediction modeling and control of evaporative condenser cooling load.
and data validation. An investigation of the total similarity Applied Thermal Engineering 25, p. 3176–3186.
of ANNs has been conducted through the use of ANN https://doi.org/10.1016/j.applthermaleng.2005.04.006
simulation. Abubaker, Ahmad M., Adnan Darwish Ahmad, Binit B. Singh,
Nelson K. Akafuah, and Kozo Saito, (2021) "Multi-objective
ABBREVIATION linear-regression-based optimization of a hybrid solar-gas
ANNs-Artificial neural networks turbine combined cycle with absorption inlet-air cooling unit."
Energy Conversion and Management 240, 114266.
TES- Thermal Energy Storage https://doi.org/10.1016/j.enconman.2021.114266
PCM-Phase Change Materials Abuella, Mohamed, and Badrul Chowdhury. (2015) "Solar power
AI-Artificial Intelligence probabilistic forecasting by using multiple linear regression
analysis." In SoutheastCon. 1–5.
BP-Back propagation https://doi.org/10.1109/SECON.2015.7132869
RMSE-Root Mean Square Error, Alhuyi Nazari, Mohammad, Mohamed Salem, Ibrahim Mahariq,
Khaled Younes, and Bashar B. Maqableh. (2021) "Utilization
MAPE-Mean Absolute Percentage Error
of data-driven methods in solar desalination systems: a
HV-Hottel-Vhillier comprehensive review." Frontiers in Energy Research 9,
SISO-single input, single output 742615. https://doi.org/10.3389/fenrg.2021.742615
Alizamir, Meysam, Sungwon Kim, Ozgur Kisi, and Mohammad
MISO-multiple input, single output
Zounemat-Kermani. (2020). "A comparative study of several
DXSAHP-direct expansion solar-aided heat pump machine learning based non-linear regression methods in
NLR-Neutrophil-Lymphocyte Ratio estimating solar radiation: Case studies of the USA and Turkey
regions”. Energy 197, 117239. https://doi.org/10.1016/
MLR-Multiple Linear Regression
j.energy.2020.117239
FFNN-feed-forward neural network Aly, A.A., Zeidan, E.B., Hamed, A.M. (2011). Performance
NARx-Nonlinear Autoregressive with exogenous input evaluation of open- cycle solar regenerator using artificial
neural network technique. Energy and Buildings 43, 454–457.
SSTES-Solar Still thermal energy systems
https://doi.org/10.1016/j.enbuild.2010.09.032
MLPNN-Multiple-layer perceptron neural network Ammar, M.B., Chaabene, M., Chtourou, Z. (2013). Artificial Neural
FFF-feed forward Network based control for PV/T panel to track optimum
thermal and electrical power. Energy Conversion and
SVR-Support Vector Regression Management 65, 372–380.
ICM-Imperia list Competition Method https://doi.org/10.1016/j.enconman.2012.08.003
GA-genetic algorithm Arcaklioglu, E., Erisen, A., Yilmaz, R. (2004). Artificial neural
network analysis of heat pumps using refrigerant mixtures.
14 VAIRAVEL MADESHWAREN et al.

Energy Conversion Management 45, p. 1917–1929. neural networks and multiple linear regression models."
https://doi.org/10.1016/j.enconman.2003.09.028 Energy Systems 11(4). 981-1002.
Bagheri, A., Esfandiari, N., Honarvar, B., & Azdarpour, A. (2020). Daut, I., M. Irwanto, Y. M. Irwan, N. Gomesh, and N. S. Ahmad.
First principles versus artificial neural network modelling of a (2011). "Combination of Hargreaves method and linear
solar desalination system with experimental validation. regression as a new method to estimate solar radiation in
Mathematical and Computer Modelling of Dynamical Perlis, Northern Malaysia." Solar Energy 85(11), 2871-2880.
Systems, 26(5). 453–480. https://doi.org/10.1016/j.solener.2011.08.026
Bagheri, A., Esfandiari, N., Honarvar, B., & Azdarpour, A. (2021). Devi, L. Gomathi, and K. Eraiah Rajashekhar, (2011). "A kinetic
ANN Modeling and experimental study of the effect of various model based on non-linear regression analysis is proposed for
factors on solar desalination. AQUA–Water Infrastructure, the degradation of phenol under UV/solar light using nitrogen
Ecosystems and Society, 70(1). 41–57. doped TiO2."Journal of Molecular Catalysis A: Chemical
Bahiraei, Mehdi, Saeed Nazari, and Habibollah Safarzadeh. (2021) 334(2), 65-76. https://doi.org/10.1016/j.molcata.
"Modeling of energy efficiency for a solar still fitted with 2010.10.025
thermoelectric modules by ANFIS and PSO-enhanced neural Dikmen, E., Ayaz, M., Ezen, H.S., Kucuksille, E.U., Sahin, A.S.
network: A nanofluid application." Powder Technology 385, (2013). Estimation and optimization of thermal performance
185-198. https://doi.org/10.1016/j.powtec.2021.03.001 of evacuated tube solar collector system. Heat Mass Transfer.
Behnam, Pooria, Meysam Faegh, and Mehdi Khiadani. (2022). "A https://doi.org/10.1007/s00231-013-1282-0
review on state-of-the-art applications of data-driven Ertunc, H.M., Hosoz, M. (2006). Artificial neural network analysis
methods in desalination systems." Desalination 532, 115744. of a refrigeration system with an evaporative condenser.
https://doi.org/10.1016/j.desal.2022.115744 Applied Thermal Engineering 26, 62–73.
Behura, A.K., Prasad, B.N., Prasad., L. (2016). Heat transfer, https://doi.org/10.1016/j.applthermaleng.2005.06.002
friction factor and thermal performance of three sides Esen, H., Esen, M., Ozsolak, O. (2015). Modelling and
artificially roughened solar air heaters. Solar Energy 130, p. experimental performance analysis of solar-assisted ground
46–59. https://doi.org/10.1016/j.solener.2016.02.006 source heat pump system. Journal of Experimental &
Behura, A.K., Prasad, B.N., Prasad., L.(2016) Investigation for heat Theoretical Artificial Intelligence, 1–17.
transfer and friction factor characteristics in three sides https://doi.org/10.1080/0952813X.2015.1056242
artificially roughened solar air heater, Ph.D. Thesis, National Esen, H., Ozgen, F., Esen, M., Sengur, A. (2009). Artificial neural
Institute of Technology, Jamshedpur, India. network and wavelet neural network approaches for
https://doi.org/10.1016/j.solener.2016.02.006 modelling of a solar air heater. Expert Systems with
Benli, H. (2013). Determination of thermal performance Applications 36, 11240–11248. http://dx.doi.org/10.1016/
calculation of two different types solar air collectors with the j.eswa.2009.02.073
use of artificial neural networks. International Journal of Heat Esmaili, Maryam, Sasan Aliniaeifard, Mahmoud Mashal, Keyvan
and Mass Transfer 60, 1–7. Asefpour Vakilian, Parisa Ghorbanzadeh, Behzad Azadegan,
https://doi.org/10.1016/j.ijheatmasstransfer.2012.12.042 Mehdi Seif, and Fardad Didaran. (2021) "Assessment of
Bhagoria J.L., Saini J.S., Solanki S.C. (2002). Heat transfer adaptive neuro-fuzzy inference system (ANFIS) to predict
coefficient and friction factor correlation for rectangular solar production and water productivity of lettuce in response to
air heater duct having transverse wedge shaped rib roughness different light intensities and CO2 concentrations."
on absorber plate, Renewable Energy 25, p. 341– Agricultural Water Management 258, 107201.
369.https://doi.org/10.1016/S0960-1481(01)00057-X https://doi.org/10.1016/j.agwat.2021.107201
Bocco, Mónica, Enrique Willington, and Mónica Arias (2010). Facão, J. , Varga, S., Oliveira, A.C. (2004). Evaluation of the Use of
"Comparison of regression and neural networks models to Artificial Neural Networks for the Simulation of Hybrid Solar
estimate solar radiation." Chilean Journal of Agricultural Collectors. International journal of green energy 1(3), 337–
Research 70(3), 428-435. https://doi.org/10.1016/ 352. https://doi.org/10.1081/GE-200033649
j.rser.2015.08.021 Faegh, Meysam, Pooria Behnam, Mohammad Behshad Shafii, and
Boesch, William W. (1982). "World's first solar powered reverse Mehdi Khiadani. (2021). "Development of artificial neural
osmosis desalination plant." Desalination 41(2), 233-237. networks for performance prediction of a heat pump assisted
https://doi.org/10.1016/S0011-9164(00)88726-3 humidification-dehumidification desalination system."
Cakmak, Gm., Yildiz, C. (2011). The Prediction of Seedy grape Desalination 508, 115052.
drying rate using a neural network method, Computers and https://doi.org/10.1016/j.desal.2021.115052
Electronics in Agriculture 75, 132–138. Farkas, I., Geczy-Vig, (2003). Neural network modeling of flat-
Caner, M., Gedik, E., Keçebas, A. (2011). Investigation on thermal plate solar collectors. Computers and Electronics in
performance calculation of two type solar air collectors using Agriculture 40, 87–102.
artificial neural network. Expert Systems with Applications 38, Fischer, S., Frey, P., Drück, H. (2012). Comparison between state-
1668–1674. https://doi.org/10.1016/j.eswa.2010.07.090 of-the-art and neural network modelling of solar collectors.
Cetiner, C., Halici, F., Cacur, H., Taymaz, I. (2005). Generating hot Solar Energy 86, 3268–3277. https://doi.org/10.1016/
water by solar energy and application of neural network. j.solener.2012.09.002
Applied Thermal Engineering 25, 1337–1348. Garud, Kunal Sandip, Simon Jayaraj, and Moo‐Yeon Lee. (2021).
https://doi.org/10.1016/j.applthermaleng.2004.09.004 "A review on modeling of solar photovoltaic systems using
Chiteka, Kudzanayi, Rajesh Arora, and S. N. Sridhara. (2020). "A artificial neural networks, fuzzy logic, genetic algorithm and
method to predict solar photovoltaic soiling using artificial hybrid models." International Journal of Energy Research
45(1). 6-35. https://doi.org/10.1002/er.5608
PREDICTIVE MODELING FOR SOLAR DESALINATION USING ARTIFICIAL NEURAL NETWORK 15

Ghermandi, Andrea, and Rami Messalem. (2009). "Solar-driven Kalogirou, S.A., Panteliou, S., Dentsoras, A. (1999). Modeling of
desalination with reverse osmosis: the state of the art." solar domestic water heating systems using artificial neural
Desalination and water treatment 7(1–3). 285-296. networks, Solar Energy 65(6), 335–342.
http://dx.doi.org/10.5004/dwt.2009.723 https://doi.org/10.1016/S0038-092X(99)00013-4
Gunasekar, N., Mohanraj, M., Velmurugan, V. (2015). Artificial Kalogirou,S.A. (2006). Prediction of flat-plate collector
neural network modeling of a photovoltaic-thermal performance parameters using artificial neural network. Solar
evaporator of solar assisted heat pumps. Energy 93, p. 908– Energ, 80, 248–259. https://doi.org/10.1016/j.solener.
922. https://doi.org/10.1016/j.energy.2015.09.078 2005.03.003
Gupta, D., Solanki, S.C., Saini, J.S. (1993). Heat and fluid flow in Kamthania, D., Tiwari, G.N. (2012). Performance Analysis of a
rectangular solar air heater ducts having transverse rib Hybrid Photovoltaic Thermal Double Pass Air Collector Using
roughness on absorber plates. Sol. Energy 51(1), 31–37. ANN. Applied Solar Energy 48(3), 186–192.
https://doi.org/10.1016/0038-092X(93)90039-Q https://doi.org/10.3103/S0003701X12030073
Hamdan, M.A., Abdelhafez, E.A., Hamdan, A.M., HajKhalil, R.A. Karabacak, K., Cetin, N. (2014). Artificial neural networks for
(2014). Heat Transfer Analysis of a Flat-Plate Solar Air controlling wind PV power systems: A review. Renewable and
Collector by Using an Artificial Neural Network. J. Infrastruct. Sustainable Energy Reviews 29, p. 804–827.
Syst, 45–78.http://dx.doi.org/10.1061/(ASCE)IS.1943- https://doi.org/10.1016/j.rser.2013.08.070
555X.0000213 Karmare, S.V. and Tikekar, A. N. (2007) Heat transfer and friction
Hosoz, M.H., Ertunc, M. (2006). Modeling of a cascade factor correlation for artificially roughened duct with metal
refrigeration system using artificial neural networks. grit ribs, Int.J.Heat MassTransfer 50, p. 4342–51.
International Journal of Energy Research 30, 12–15. Khanna, Vandana, B. K. Das, Dinesh Bisht, and P. K. Singh. (2015).
https://doi.org/10.1002/er.1218 "A three diode model for industrial solar cells and estimation
Ibrahim, S., I. Daut, Y. M. Irwan, M. Irwanto, N. Gomesh, and Z. of solar cell parameters using PSO algorithm." Renewable
Farhana. (2012). "Linear regression model in estimating solar Energy 78: 105–113. https://doi.org/10.1016/
radiation in Perlis." Energy Procedia 18: 1402–1412. j.renene.2014.12.072
https://doi.org/10.1016/j.egypro.2012.05.156 Kharb, Ravinder Kumar, S. L. Shimi, S. Chatterji, and Md Fahim
Jani, D.B., Mishra, M., Sahoo, P.K. (2017). Application of artificial Ansari. (2014) "Modeling of solar PV module and maximum
neural network for predicting performance of solid desiccant power point tracking using ANFIS." Renewable and
cooling systems A review. Renewable and Sustainable Energy Sustainable Energy Reviews 33: 602-612.
Reviews 80, p. 352–366. https://doi.org/10.1016/ https://doi.org/10.1016/j.rser.2014.02.014
j.rser.2017.05.169 Khare, Anula, and Saroj Rangnekar. (2013). "A review of particle
Jaurker, A.R., Saini, J.S., and Gandhi, B.K..(2006). Heat transfer and swarm optimization and its applications in solar photovoltaic
friction characteristics of rectangular solar air heater duct system." Applied Soft Computing 13(5), 2997-3006.
using artificial grooved roughness, Solar Energy., 80, p. 895 https://doi.org/10.1016/j.asoc.2012.11.033
907. https://doi.org/10.1016/j.solener.2005.08.006 Kicsiny, Richárd. (2014). "Multiple linear regression based model
Jervase, Joseph A., Hadj Bourdoucen, and Ali Al-Lawati. (2001). for solar collectors." Solar Energy 110, 496-506.
"Solar cell parameter extraction using genetic algorithms." Köysal, Yavuz, Ali Ekber Özdemir, and Tahsin Atalay.
Measurement science and technology 12(11). 1922. "Experimental and modeling study on solar system using
https://doi.org/10.1088/0957-0233/12/11/322 linear fresnel lens and thermoelectric module." Journal of
Kalogirou, S.A. (2000). Applications of artificial neural-networks Solar Energy Engineering” 140(6): 061003.
for energy systems. Applied Energy 67, p. 17–35. https://doi.org/10.1115/1.4039777
https://doi.org/10.1016/S0306-2619(00)00005-2 Layek, A., Saini, J. S., Solanki, S. C., (2009). Effect of chamfering on
Kalogirou, S.A. (2000). Long-term performance prediction of heat transfer and friction characteristics of solar air heater
forced circulation solar domestic water heating systems using having absorber plate roughened with compound turbulators,
artificial neural networks. Applied Energy 66, p. 63–74. Renewable Energy 34, p. 1292–1298.
https://doi.org/10.1016/S0306-2619(99)00042-2 https://doi.org/10.1016/j.renene.2008.09.016
Kalogirou, S.A. (2001). Artificial neural networks in renewable Lecoeuche, S., Lalot, S. (2005). Prediction of the daily performance
energy systems applications: a review. Renewable and of solar collectors. International Communications in Heat and
Sustainable Energy Reviews 5, p. 373–401. Mass Transfer 32, 603–611.
https://doi.org/10.1016/S1364-0321(01)00006-5 https://doi.org/10.1016/j.icheatmasstransfer.2004.12.002
Kalogirou, S.A., Mathioulakis, E., Belessiotis, V. (2014). Artificial Mahesh, P. Venkata, S. Meyyappan, and Rama Koteswara Rao
neural networks for the performance prediction of large solar Alla. (2022). "A new multivariate linear regression MPPT
systems. Renewable Energy 63, p. 90–97. algorithm for solar PV system with boost converter." ECTI
https://doi.org/10.1016/j.renene.2013.08.049 Transactions on Electrical Engineering, Electronics, and
Kalogirou, S.A., Neocleous, C.C., Schizas, C.N. (1998). Artificial Communications 20(2), 269-281. https://doi.org/10.37936/
neural networks for modelling the starting-up of a solar ecti-eec.2022202.246909
steam-generator, Applied Energy, 60, 89–100. Manolakos, D., Essam Sh Mohamed, I. Karagiannis, and G.
Kalogirou, S.A., Panteliou, S., Dentsoras, A. (1999). Artificial neural Papadakis. (2008). "Technical and economic comparison
networks used for the performance prediction of a between PV-RO system and RO-Solar Rankine system. Case
thermosiphon solar water heater, Renewable Energy, 18, 87– study: Thirasia island." Desalination 221(1-3). 37-46.
99. https://doi.org/10.1016/S0960-1481(98)00787-3 https://doi.org/10.1016/j.desal.2007.01.066
16 VAIRAVEL MADESHWAREN et al.

Mashaly, A. F., & Alazba, A. A. (2016). Comparison of ANN, MVR, Prasad, R.K., Sahu, M.K. (2017). Entropy generation and
and SWR models for computing thermal efficiency of a solar thermodynamic analysis of solar air heaters with artificial
still. International Journal of Green Energy, 1016–1025. roughness on absorber plate, Archives of Thermodynamics 38
Mashaly, Ahmed F., and A. A. Alazba. (2017). "Application of , p. 23–48, DOI: 10.1515/aoter-2017-0014.
adaptive neuro-fuzzy inference system (ANFIS) for modeling Prasad,R.K., Sahu, M.K. (2016). Investigation on Optimal
solar still productivity." Journal of Water Supply: Research and Thermohydraulic Performance of a Solar Air Heater Having
Technology—AQUA 66(6). 367–380. https://doi.org/10.2166/ Arc Shaped Wire Rib Roughness on Absorber Plate,
aqua.2017.138 International Journal of Thermodynamics 19, p. 214–224.
Mashaly, Ahmed F., and A. A. Alazba. (2018) "ANFIS modeling and http://dx.doi.org/10.5541/eoguijt.277772
sensitivity analysis for estimating solar still productivity using Ramazanian, Saeed, Mehdi Aliehyaei, Mahmoud Salimi, and
measured operational and meteorological parameters." Mohammad Mehdi Najafizadeh. (2023). "Theoretical and
Water Science and Technology: Water Supply 18(4), 1437– experimental investigation of reverse osmosis (RO)
1448. https://doi.org/10.2166/ws.2017.208 desalination solar system using the solar panel, battery, and
Mellit, A., Kalogirou, S.A., Hontoria, L., Shaari, S. (2009). Artificial water turbine for high-pressure recovery of output brine."
intelligence techniques for sizing photovoltaic systems: a Solar Energy 262, 111852. http://dx.doi.org/
review. Renewable and Sustainable Energy Reviews 13, p. 10.1016/j.solener.2023.111852
406 19. https://doi.org/10.1016/j.rser.2008.01.006 Ramedani, Zeynab, Mahmoud Omid, Alireza Keyhani, Benyamin
Mohanraj, M., Jayaraj, S., Muraleedharan, C. (2008). Modeling of Khoshnevisan, and Hadi Saboohi. (2014). "A comparative
a direct expansion solar assisted heat pump using artificial study between fuzzy linear regression and support vector
neural networks. International Journal of Green Energy 5, regression for global solar radiation prediction in Iran." Solar
520–532. https://doi.org/10.1080/15435070802498499 Energy 109, 135-143. https://doi.org/10.1016/
Mohanraj, M., Jayaraj, S., Muraleedharan, C. (2009). Exergy j.solener.2014.08.023
analysis of direct expansion solar-assisted heat pumps using Rezk, Hegazy, Ramy H. Mohammed, Emanne Rashad, and Ahmed
artificial neural networks. Int. J. Energy Res. 33, 1005–1020. M. Nassef. (2022). "ANFIS-based accurate modeling of silica
https://doi.org/10.1002/er.1534 gel adsorption cooling cycle." Sustainable Energy
Mohanraj, M., Jayaraj, S., Muraleedharan, C. (2009). Performance Technologies and Assessments 50, 101793.
prediction of a direct expansion solar assisted heat pump https://doi.org/10.1016/j.seta.2021.101793
using artificial neural networks. Applied Energy 86, 1442– Ruivo, Celestino Rodrigues, Xabier Apaolaza-Pagoaga, Giovanni Di
1449. https://doi.org/10.1016/j.apenergy.2009.01.001 Nicola, and Antonio Carrillo-Andrés. (2022). "On the use of
Mohanraja, M., Jayaraj, S., Muraleedharan, C. (2012). experimental measured data to derive the linear regression
Applications of artificial neural networks for refrigeration, air- usually adopted for determining the performance parameters
conditioning and heat pump systems A review. Renewable of a solar cooker." Renewable Energy 181, 105-115.
and Sustainable Energy Reviews 16, p. 1340–1358. https://doi.org/10.1016/j.renene.2021.09.047
https://doi.org/10.1016/j.rser.2011.10.015 Saadaoui, Driss, Mustapha Elyaqouti, Khalid Assalaou, and Souad
Nassef, Ahmed M., Shek Md Atiqure Rahman, Hegazy Rezk, and Lidaighbi. (2021). "Parameters optimization of solar PV
Zafar Said. (2020). "ANFIS-based modelling and optimal cell/module using genetic algorithm based on non-uniform
operating parameter determination to enhance cocoa beans mutation." Energy Conversion and Management: X 12,
drying-rate." IEEE Access 8, 45964–45973. 100129. https://doi.org/10.1016/j.ecmx.2021.100129
Olabi, A. G., Hegazy Rezk, Enas Taha Sayed, Rania M. Ghoniem,
Sahu, M.K., Prasad, R. K. (2016). ENergy based performance
and Mohammad Ali Abdelkareem. (2023). "Boosting carbon
evaluation of solarair heater with arcs haped wire roughened
dioxide adsorption capacity applying Jellyfish optimization
absorber plate, Renewable Energy 96, p. 233–243.
and ANFIS-based modelling." Ain Shams Engineering Journal
https://doi.org/10.1016/j.renene.2016.04.083
14(4), 101931. https://doi.org/10.1016/j.asej.2022.101931
Sahu, M.K., Prasad, R. K. (2017). Thermohydraulic performance
Pacheco-Vega, A., Sen, M., Yang, K.T., McClain, R.L. (2001). Neural
analysis of an arc shape wire roughened solar air heater,
network analysis of an-tube refrigerating heat exchanger with
Renewable Energy 108, p. 598–614
limited experimental data. International Journal of Heat and
https://doi.org/10.1016/j.renene.2017.02.075
Mass Transfer 44, p. 763–70. https://doi.org/10.1016/S0017-
Saini, S.K., Saini, R.P. (2008). Development of correlations for
9310(00)00139-3
Nusselt number and friction factor for solar air heater with
Perendeci, Altunay, Sever Arslan, Serdar S. Çelebi, and
roughened duct having arc-shaped wire as artificial
Abdurrahman Tanyolaç. (2008) "Prediction of effluent quality
roughness. Solar Energy 82, p. 1118–3110.
of an anaerobic treatment plant under unsteady state
https://doi.org/10.1016/j.solener.2008.05.010
through ANFIS modeling with on-line input variables."
Chemical Engineering Journal 145(1), 78–85. Salem, H., Kabeel, A. E., El-Said, E. M., & Elzeki, O. M. (2022).
https://doi.org/10.1016/j.cej.2008.03.008 Predictive modelling for solar power-driven hybrid
Pérez, José Antonio, Manuel González, and Daniel Dopico. (2010) desalination system using artificial neural network regression
"Adaptive neurofuzzy ANFIS modeling of laser surface with Adam optimization. Desalination, 522, 115411.
treatments." Neural Computing and Applications 19: 85–90. Singh, A.P., Siddhartha, V. (2014). Effect of artificial roughness on
Prasad, B.N., Saini, J.S. (1988). Effect of artificial roughness on heat transferand friction characteristics having multiple arc
heat transfer and friction factor in a solar air heater. Solar shaped roughness element on the absorber plate, Solar
energy 6, 555–560. https://doi.org/10.1016/0038- Energy 105, p. 479–493 https://doi.org/10.1016/
092X(88)90058-8 j.solener.2014.04.007
PREDICTIVE MODELING FOR SOLAR DESALINATION USING ARTIFICIAL NEURAL NETWORK 17

Skumanich, A., Cr Smythe, and E. N. Frazier. (1975) "On the Transfer 33, p. 898–907. https://doi.org/10.1016/
statistical description of inhomogeneities in the quiet solar j.icheatmasstransfer.2006.04.003
atmosphere. I-Linear regression analysis and absolute Yilmaz, S., Atik, K. (2007). Modeling of a mechanical cooling
calibration of multichannel observations of the Ca/+/emission system with variable cooling capacity by using artificial neural
network." Astrophysical Journal, vol. 200 (1), 747-764. network. Applied Thermal Engineering 27, p. 2308–2313.
https://ui.adsabs.harvard.edu/link_gateway/1975ApJ...200.. https://doi.org/10.1016/j.applthermaleng.2007.01.030
747S/doi:10.1086/153846 Zahraee, S.M., Assadi, M.K. , Saidur, R. (2016). Application of
Sohani, A., Hoseinzadeh, S., Samiezadeh, S., & Verhaert, I. (2021). Artificial Intelligence Methods for Hybrid Energy System
Machine learning prediction approach for dynamic Optimization. Renewable and Sustainable Energy Reviews 66,
performance modeling of an enhanced solar still desalination p. 617–630. https://doi.org/10.1016/j.rser.2016.08.028
system. Journal of Thermal Analysis and Calorimetry, 1–12. Zayed, Mohamed E., Jun Zhao, Wenjia Li, Ammar H. Elsheikh, and
Soleimanzade, Mohammad Amin, Amit Kumar, and Mohtada Mohamed Abd Elaziz. (2021) "A hybrid adaptive neuro-fuzzy
Sadrzadeh. (2022). "Novel data-driven energy management inference system integrated with equilibrium optimizer
of a hybrid photovoltaic-reverse osmosis desalination system algorithm for predicting the energetic performance of solar
using deep reinforcement learning." Applied Energy 317, dish collector." Energy 235, 121289.
119184. https://doi.org/10.1016/j.apenergy.2022.119184 https://doi.org/10.1016/j.energy.2021.121289
Sözen, A., Menlik, T., Ünvar, S. (2008). Determination of efficiency
of flat plate solar collector using neural network. Expert
Systems with Applications 35, 1533–1539.
https://doi.org/10.1016/j.eswa.2007.08.080
Tripathy, P.P., Kumar, S. (2009). Neural network approach for
food temperature prediction during solar drying.
International Journal of Thermal Sciences 48, 1452–1459.
https://doi.org/10.1016/j.ijthermalsci.2008.11.014
Uma Maheswari, C., and R. Meenakshi Reddy. 2018, “CFD Analysis
of Different Types of Single Basin Solar Stills.” IOP Conference
Series: Materials Science and Engineering 330 (1): 012097.
Varol, Y., Koca, A., Oztop, H.F., Avci, E. (2010). Forecasting of
thermal energy storage performance of Phase Change
Material in a solar collector using soft computing techniques.
Expert Systems with Applications 37, 2722–2732.
https://doi.org/10.1016/j.eswa.2009.08.007
Verma, S.K., Prasad, B.N. (2000). Investigation for the optimal
thermo- hydraulic performance of artificially roughened solar
air heaters. Renewable Energy 20, p. 19–36.
https://doi.org/10.1016/S0960-1481(99)00081-6
Wahbah, Maisam, Samuel Feng, Tarek HM EL-Fouly, and Bashar
Zahawi. (2019). "Root-transformed local linear regression for
solar irradiance probability density estimation." IEEE
Transactions on Power Systems 35(1), 652–661.
https://doi.org/10.1109/TPWRS.2019.2930699
Wang, L., Kisi, O., Kermani, M.Z., Salazar, G.A., Zhu, Z., Gong, W.
(2016). Solar radiation prediction using different techniques:
model evaluation and comparison, Renewable and
Sustainable Energy Reviews 61, p. 384–397.
https://doi.org/10.1016/j.rser.2016.04.024
Xie, H., Liu, L., Ma, F., Fan, H. (2009). Performance prediction of
solar collectors using artificial neural networks. International
Conference on Artificial Intelligence and Computational
Intelligence, 573–576. https://doi.org/10.1109/AICI.2009.344
Yadav, A.K., Chandel, S.S. (2014). Solar radiation prediction using
Artificial Neural Network techniques: A review. Renewable
and Sustainable Energy Reviews 33, p. 772–781.
https://doi.org/10.1016/j.rser.2013.08.055
Yaïci, W., Entchev, E. (2014). Performance prediction of a solar
thermal energy system using artificial neural networks.
Applied Thermal Engineering 73, p. 1348–1359.
https://doi.org/10.1016/j.applthermaleng.2014.07.040
Yigit, K.S., Ertunc, H.M. (2006). Prediction of the air temperature
and humidity at the outlet of a cooling coil using neural
networks. International Communications in Heat and Mass