Solar Energy 224 (2021) 500–515

Contents lists available at ScienceDirect

Solar Energy
journal homepage: www.elsevier.com/locate/solener

Review of application of AI techniques to Solar Tower Systems

Kypros Milidonis a, *, Manuel J. Blanco a, Victor Grigoriev a, Constantinos F. Panagiotou c,
Aristides M. Bonanos a, Marios Constantinou a, John Pye b, Charles-Alexis Asselineau b
a
Energy Environment and Water Research Center, The Cyprus Institute, Nicosia 2121, Cyprus
b
Research School of Electrical, Energy and Materials Engineering, The Australian National University, Canberra, Australia
c
Eratosthenes Centre of Excellence, Cyprus University of Technology, Limassol 3036, Cyprus

A R T I C L E I N F O A B S T R A C T

Keywords: Artificial Intelligence (AI) is increasingly playing a significant role in the design and optimization of renewable
Concentrating solar thermal energy systems. Many AI approaches and technologies are already widely deployed in the energy sector in ap­
Solar towers plications such as generation forecasting, energy efficiency monitoring, energy storage, and overall design of
Central receiver systems
energy systems. This paper provides a review of the applications of key AI techniques on the analysis, design,
Artificial intelligence
optimization, control, operation, and maintenance of Solar Tower systems, one of the most important types of
Optimization
Metaheuristics Concentrating Solar Thermal (CST) systems. First, key AI techniques are briefly described and relevant examples
Artificial neural networks of their application to CST systems in general are provided. Subsequently, a detailed review of how these AI
techniques are being used to advance the state of the art of solar tower systems is presented. The review is
structured around the different subsystems of a solar tower system.

techniques may play in the renewable energy sector. In 2010, Kalogirou

and Sencan (2010) performed a comprehensive review on how AI
1. Introduction
techniques can be applied to improve different aspects related to the
functioning of diverse solar energy systems. More recently, Belu (2014)
Our societies are constantly being transformed. One of the most
provided an overview of commonly used AI methodologies in solar en­
relevant transformations they are currently undergoing is increased
ergy, paying special attention to the use of neural networks, fuzzy logic,
digitization. This transformation has already led to novel business
and genetic algorithms, while Youssef et al. (2017) investigated the
models and services and is progressing towards a dematerialized data
application of AI techniques to the design and control of solar energy
economy, with digitally provided services increasingly in the foreground
systems. Focusing on solar based energy systems and devices, H.
(Liggesmeyer et al., 2019). Integration of ”smart systems” with infor­
Elsheikh et al. (180:), performed a comprehensive review on employing
mation technology platforms provides the infrastructure for connecting
Artificial Neural Networks (ANNs) for their optimization, while Oliva
digital representations of things (”digital twins”) to physical objects
et al. (2019) provided a thorough review on the use of meta-heuristic
(Internet of Things) (Liggesmeyer et al., 2019).
approaches for enhancing the efficiency of solar cells. These analyses,
Due to the nature and scales of the infrastructures involved, the
however, have been very broad in scope, and necessarily not very deep
magnitude of the investments, and associated legal frameworks and
in the treatment of specific renewable energy technologies. The
regulations, the increased digitization of the energy sector may take
approach of the present article is almost the opposite, we want to
place at a slower pace than other sectors, but it is already underway. In
analyze in depth the applications of AI techniques to advance the state of
this transformation, Artificial Intelligence (AI) is playing a relevant role.
the art of a very particular type or solar energy system: the solar tower or
The employment of AI techniques in renewable energy has already
central receiver system.
proven effective, having a major impact on addressing challenges within
The solar tower is one of the most promising types of Concentrating
the sector (Jha et al., 2017). Currently, the vast majority of AI applica­
Solar Thermal (CST) systems. They can provide dispatchable electricity
tions in the renewable energy field focus on improving energy resource
and high-temperature process heat thanks to their capability of
forecasting, predictive maintenance of operating components and de­
achieving very-high operating temperatures and integrating large ther­
mand prediction.
mal energy storage in a very cost-effective way. Furthermore, they have
Several attempts have been made to analyze the role that AI

* Corresponding author.
E-mail address: [email protected] (K. Milidonis).

https://doi.org/10.1016/j.solener.2021.06.009
Received 1 December 2020; Received in revised form 24 May 2021; Accepted 2 June 2021
Available online 19 June 2021
0038-092X/© 2021 International Solar Energy Society. Published by Elsevier Ltd. All rights reserved.
K. Milidonis et al. Solar Energy 224 (2021) 500–515

Nomenclature HTF Heat Transfer Fluid

KHO Krill Herd Optimization
ABC Artificial Bee Colony LCOE Levelized Cost of Electricity
ACO Ant Colony Optimization LM Levenberg–Marquardt
AI Artificial Intelligence MBPC Model Based Predictive Control
ANFIS Artificial Neural Network and Fuzzy Inference Engine MCRT Monte Carlo Ray-Tracing
ANN Artificial Neural Network mGA Micro Genetic Algorithm
AV Artificial Vision MLP Multilayer Perceptron
BP Back Propagation MPC Model Predictive Control
BPNN Back Propagation Neural Network NN Neural Network
CMAC Cerebellar Model Arithmetic Computer NSGA Non-dominated Sorting Genetic Algorithm
CSP Concentrated Solar Power PSO Particle Swarm Optimization
CST Concentrated Solar Thermal RBF Radial Basis Function
dCNN deep Convolution Neuronal Network RBFN Radial Basis Function Network
DEPSO Differential Evolution PSO RHONN Recursive High Order Neural Networks
DL Deep Learning SA Simulated Annealing
DSE-DE Dynamic Dpeciation-based Mutation Differential SCADA Supervisory Control and Data Acquisition
Evolution SCG Scaled Conjugate Gradient
EA Evolutionary Algorithm SMC Sliding Modes Control
EKF Extended Kalman Filter SOA Simulation–Optimization Algorithms
F-PI Fuzzy Proportional–Integral controller SPSO Stochastic PSO
F-PID Fuzzy Proportional–Integral–Derivative controller TES Thermal Energy Storage
FL Fuzzy Logic TOPSIS Technique for Order Performance by Similarity to Ideal
FPGA Field Programmable Gate Array Solution
GA Genetic Algorithm

experienced dramatic cost reductions (47%) between 2010 and 2019 • The emerging area of multi-tower heliostat field and receiver system
(IRENA, 2020). Other types are the parabolic trough, the Linear Fresnel optimization (3.3)
and the parabolic dish. The main components of a solar tower system are • The maintenance and diagnosis on such systems (3.5)
summarized in Fig. 1. In it, a field of mirrors, called ”heliostats”, are • The optimization of hybrid solar tower and photovoltaic systems
tracking the sun in such a way as to always reflect the direct sunlight (3.6)
towards a receiver that is typically located atop of a tower. Typically, at • The overarching system design and optimization (3.7)
the receiver, the concentrated direct solar radiation is transformed into
the enthalpy increase of a heat transfer fluid (HTF) and this, in turn, is With regards to the above applications, this review paper considers
used to run a cyclic thermal process, such as a conventional power block. all AI methods and techniques found in the literature including: Machine
This review paper focuses on identifying the main AI techniques used Learning (ML), Artificial Neural Networks (ANN), Fuzzy Logic (FL),
in this field and discusses their application to five key topics: metaheuristics such as Evolutionary Algorithms (EA), as well as hybrid
combinations between two or more of the above techniques. In addition,
• The design of the heliostat field (3.2), the receiver system (3.1) and relevant papers referring to combinations of classical optimization al­
the thermal storage system (3.4) gorithms with one of the above AI branches are also considered here.
Although there is no widely accepted definition of AI (Wang, 2019), we

Fig. 1. Typical layout of a Concentrated Solar Tower Plant.

501
K. Milidonis et al. Solar Energy 224 (2021) 500–515

consider as AI the above-mentioned methods as they differ from deter­ individuals in a population of evaluations. In analogy to natural selec­
ministic numerical optimisation by at least one feature: they are agent- tion processes, the least ”fit” members of a population are discarded,
based models; they learn from training or simulated data to modify their whereas the ”fit” members evolve through different operations until the
strategy to reach a solution or they involve stochastic decision processes, ”fittest” members of the population are found. Here, a brief introduction
independent of inputs uncertainty, to reach a solution. is given for two of the major evolutionary algorithms that are used in
Finally, there is a very relevant set of applications of AI techniques CST plants (i.e. Genetic Algorithms (GA) and Particle Swarm Optimi­
that benefit the operation of solar towers and other CST systems, which zation (PSO)), while additional ones will be discussed in the next section
is not addressed in this article, however. It is the set related to the short, (i.e. Ant Colony Optimization (ACO), Krill Herd Optimization (KHO),
mid and long term solar resource forecasting. This article does not Artificial Bee Colony (ABC), etc).
address this, because it is in a class of its own and has already being
treated abundantly by others, i.e. (Dobbs et al., 2017; Mubiru and 2.3.1. Genetic Algorithms
Banda, 2008; Yadav and Chandel, 2014; Behrang et al., 2010; Cha­ Genetic algorithms imitate the biological mechanisms of reproduc­
turvedi and Isha, 2016; Mellit, 2008). tion and natural selection. First, an initial population of candidates (i.e.
parameter sets) is arbitrarily chosen for the problem. The ”fitness” of
2. A general overview of Artificial intelligence techniques in each candidate is evaluated by the fitness function and the best solutions
Concentrated Solar Thermal Systems are chosen to generate the next population of solutions using specific
heuristics: crossover and mutation operators. The crossover operator
This section provides a classification for a range of AI methods that resembles the process by which two chromosomes pair-up and exchange
are mostly used in CST systems, including Concentrating Solar thermal portions of their DNA. The mutation operator is used to maintain the
Power (CSP) plants, and gives a short introduction of the techniques by diversity in the population by changing some genes in the DNA
briefly exploring their application on the optimization of CSP Plants. sequence. The above process is repeated until a predefined termination
criterion is satisfied, or until the maximum number of iterations is
2.1. Artificial Neural Network achieved. GAs typically converge more slowly than other more tradi­
tional optimization approaches but they are powerful since they can
The emergence of ANN methods is motivated by their similarity to provide optimal solutions to problems that lack continuity, linearity,
successfully working biological systems, consisting of very simple but convexity, derivatives, etc, which is typical in CST systems. For instance,
numerous nerve cells that work in massively parallel arrangements and, Cabello et al. (2011) proposed a simplified GA-based model in order to
most importantly, are capable of learning from training samples. As a optimize the size of a parabolic trough collector for maximum yearly
result, this technique can provide solutions to highly non-linear prob­ profit, motivated by the inability of conventional methods to provide
lems in multi-dimensional information domains in a fast and simple accurate predictions. The algorithm was also used to minimize costs and
manner, thus making ANN methods widely used in various problems, losses associated with parabolic dish based solar power plants (Cump­
including concentrating solar thermal configurations. For instance, May ston and Pye, 2015).
et al. (2016) successfully utilized an ANN inverse model to estimate the
thermal performance in a parabolic trough. Fischer et al. (2012) 2.3.2. Particle Swarm Optimization
demonstrated that ANNs achieve better agreement than state-of-the-art Particle swarm optimization (PSO) methods mimic the behavior of
conventional approaches for the thermal performance of the collector bird flocks scavenging for food, which optimally cover space and make
field. Ghritlahre and Prasad (2018) conducted a review work in order to use of collective intelligence to maximize the efficiency of the process. In
assess the ability of ANN on predicting the performance of solar energy PSO, the objective function evaluations are considered as particles in a
collectors more accurately than conventional models. multi-dimensional parameter space. The objective is to determine the
optimal position within the parameter space using a population (or
2.2. Fuzzy Logic swarm) of particles subject to specific movement rules. The optimal
positions in the parameter space are the optimal sets of parameters able
Fuzzy logic is a specific branch of logic in which the validity of to give the best evaluation of the objective function. The swarm posi­
statements can have a ”degree of truth” rather than an absolute true or tions are updated using the velocity of each particle and the shared in­
false answer, as is found in Boolean logic. As such, FL imitates human formation between the particles. The benefit of using PSO instead of
decision-making. For system description, FL relies on conditional ”if- other AI techniques, such as GA, is that it requires less optimization
then” rules that are used to characterize how components interact. This specific parameters and is consequently less computationally expensive.
approach is elegant due to its simplicity and flexibility, thus allowing to This technique has appeared in several research works involving
handle real-world problems with imprecise and incomplete information. different CST system problems. Farges et al. (2013) optimised the annual
In CST technology, this methodology has been mainly used on control heliostat field optical efficiency and the thermal energy stored in a solar
problems. An illustrative example is the work done by Rubio et al. tower system by blending a PSO and a Monte-Carlo algorithm. Cheng
(1995) who applied fuzzy logic to a distributed collector field in order to et al. (2015) used a similar technique to assess the optical performance
control the temperature of the fluid that exits the receiver tube, while a of parabolic trough solar collector systems. A summary of various ap­
more recent example is the work conducted by Xu et al. (2017) who plications of PSO in solar energy systems can be found in Elsheikh and
developed a control strategy of outlet steam temperature in a Direct Elaziz (2018).
Steam Generation (DSG) parabolic trough plant.
2.4. Simulated Annealing
2.3. Evolutionary Algorithms
Simulated annealing (SA) are stochastic techniques inspired by
Evolutionary algorithms (EA) are computational stochastic methods thermodynamics and they are used to find approximate global optima of
that resemble biological processes and are used to solve complex prob­ given functions. These type of algorithms are also used to provide an
lems. Central to EA is the concept of ”fitness”, which is used to determine efficient exploration of the parameter space in search of possible mul­
how efficiently an evaluation of the objective function (or fitness func­ tiple local minima. SA techniques have found different applications in
tion) is able to perform with regards to the objective function minimi­ CST, such as to optimize the receiver tube of a parabolic trough that uses
zation (or maximization). Evolutionary algorithms are stochastic water as working fluid (Nolte et al., 2013), and the optical optimization
methods in which objective function evaluations are treated as of a Linear Fresnel collector as a function of the mirror width and focal

502
K. Milidonis et al. Solar Energy 224 (2021) 500–515

lengths (Boito and Grena, 2016). • Receivers are generally impossible to simulate purely analytically
and require numerical methods to handle complex, often non-
2.5. Hybrid Techniques differentiable, functions.
• The evaluation of the receiver performance generally involves sto­
Hybrid techniques aim to integrate different AI algorithms in ways chastic methods such as Monte-Carlo ray-tracing, giving rise to un­
that overcome individual limitations and achieve synergistic effects certainty in the results which can be very complex for optimization
through hybridization of these algorithms, thus leading to better opti­ methods to adapt to.
mization performance and accuracy. In this section, a brief introduction
will be given for the most commonly used algorithms in CST, while more In this review, receivers are classified into three distinct categories
specific implementations are discussed in Section 3. depending on their general characteristics. Cavity receivers have a
concave geometry that tend to reduce thermal losses and, therefore,
2.5.1. Adaptive Neuro-Fuzzy Inference increase the light-to-heat conversion efficiency. External receivers have
Adaptive Neuro-Fuzzy Inference (ANFIS) is a hybrid intelligent a vertical cylindrical or prismatic geometry that receive concentrated
method that exploits both the learning ability of the ANN and the radiation form a surrounding heliostat field. Volumetric receivers are
knowledge representation of fuzzy logic in order to handle complex non- composed of porous materials that concentrated sunlight penetrates and
linear and dynamic systems. This technique combines the components of through which a heat transfer fluid is circulated. The techniques used for
a conventional fuzzy system with a hidden ANN layer to perform com­ tower receiver optimization are summarized in Table 1. (see Table 2).
putations. The fuzzy logic component provides flexibility and adapt­
ability, while the neural network’s learning capacity is used to further 3.1.1. Cavity receivers
enhance the system knowledge and improve the computation efficiency. Asselineau et al. (2015) employed a stochastic optimization method
Yaici and Entchev (2016) proposed an ANFIS model to predict the considering complex coupled configurations and design constraints
performance of a solar thermal energy system, described by several along with simple heuristics for optimizing the geometry of cavity re­
climatic and thermo-hydraulic design variables. Their results showed ceivers based on their optical and thermal efficiency. They concluded
that ANFIS can provide high accuracy and reliability for predicting the that the radiative losses and heat transfer efficiency are influenced by
performance of this type of energy systems. Another application of the receiver shape, leading thus to receiver geometries that balance the
ANFIS model was discussed by Mohammadi et al. (2015) who demon­ heat-loss reduction with adequate heat transfer to the working fluid. The
strated the ability of this technique to provide estimations of the daily same team (Asselineau et al., 2015) also presented a stochastic optimi­
horizontal global solar radiation with favorable level of reliability and zation method which involved Monte-Carlo Ray-Tracing (MCRT) and
precision. adaptations to illumination and radiative heat transfer problems, and
implemented the approach on solar receiver geometry optimization.
2.5.2. Hybrid Genetic Algorithm-Particle Swarm Optimization Their objective was to progressively screen the best-performing receiver
Although Genetic Algorithms (GAs) have proven to be robust and geometries in a population, discarding under-performing candidates as
effective in solving a range of engineering optimization problems, there soon as they were identified in order to minimize the computational cost
are some drawbacks that prohibit their application on other problems. of the optimization process. To homogenize the solar flux distribution in
The high implementation cost and the requirement for high number of a cavity receiver while keeping the optical loss as low as possible, Wang
iterations for convergence are considered as the two major drawbacks. et al. (2017) performed a multi-objective optimization using a non-
On the other hand, although the strength of the PSO algorithms is the dominated genetic algorithm (NSGA) that optimizes the solar absorp­
fast convergence, this convergence is often premature. Hybrid Genetic tivity distribution within the receiver. These authors concluded that an
Algorithm-Particle Swarm Optimization (GA-PSO) methods, are algo­ optimal Pareto front can be obtained with the multi-objective optimi­
rithms that combine the advantages of both in order to achieve fast zation, which can find the best compromise between the reflection losses
convergence along with a high capacity of global convergence (Sharma and the non-uniformity of the solar flux distribution. The optimization
and Shankar Singhal, 2015). GA-PSO algorithm runs the refinement
process of the solutions from PSO within multiple populations, leading Table 1
to a suitable method to optimize large domain objective functions, Summary of algorithms the considered in this review for tower receiver geom­
applicable in a wide range of problems. For example, Senjyu et al. etry optimization.
(2008) adopted this technique in order to solve a unit commitment Author Ref. Year Algorithm
problem for thermal units integrated with wind and solar energy sys­
External receivers
tems. They successfully demonstrated the effectiveness of this algorithm
Asselineau (Asselineau et al., 2016; 2016 Multi-objective with
compared to several established methods for thermal power system of et al. Asselineau et al., 2017) stochastic
various sizes, leading to high quality (low production cost) solutions. optimization and
machine learning
3. Artificial intelligence techniques in concentrated solar Luo et al. (Luo et al., 2017) 2017 NSGA
Cavity receivers
thermal tower systems Daun et al. (Daun et al., 2003; Daun 2003 Gradient based
et al., 2003) optimization
3.1. Receiver geometry optimization Sarvani (Sarvari, 2007) 2007 GA
Farahmand (Farahmand and Payan, 2012 PSO
et al. 2012)
The geometry of receivers is a key factor that influences their per­
Asselineau (Asselineau et al., 2015; 2015 Stochastic optim. &
formance and reliability. Receiver design is a complex multi-variate, et al. Asselineau et al., 2015) heuristics
multi-physics and multi-objective problem. Receiver optimization is Qi et al. (Qi et al., 2016) 2016 PSO
difficult because of the following: Sun et al. (Sun et al., 2016) 2016 KHO
Wang et al. (Wang et al., 2017) 2017 Non dominated GA
Schöttl et al. (Schöttl et al., 2020) 2020 EA, ANN
• The objective functions are generally highly non-linear functions of Darvishvand (Darvishvand et al., 2016) 2015 GA
the geometrical parameters which prevents the use of simple et al.
gradient descent based methods. Volumetric receivers
Du et al. (Du et al., 2017; Du et al., 2017/ GA
2018) 8

503
K. Milidonis et al. Solar Energy 224 (2021) 500–515

Table 2
Summary of algorithms used in field layout optimization.
Algorithm Count Algorithm Count

GA 8 Greedy heuristic 2
PSO 4 Differential evolution 3
ABC 1 SA 1
PSO + GA 2

process resulted into a significant improvement of the flux distribution

uniformity within the cavity and led to an optimized solar energy dis­
tribution with increased view factors to the cavity aperture. Recently,
Schöttl et al. (2020) employed an EA in order to find the global optimum
of a multi-parametric cavity receiver design for achieving an optical to
thermohydraulic interface that would assure maximum yield. The free
parameters, which were optimized simultaneously, included aspects
related to the receiver geometry coupled with characteristics of the
aiming strategy and receiver hydraulic circuit. The optimization process
was accelerated using an ANN to solve the receiver heat transfer prob­
lem. The ANN was trained on data from a more complex model and
acted as a surrogate mode. The authors reported that their approach
yielded more than 4% increment in terms of the annual yield when
compared to the reference configuration.
Cavity receivers are radiant enclosures. Within the literature, few
papers exist regarding the use of evolutionary algorithms for the opti­
mization of such enclosures, often with specular surfaces, but are not
directly related to solar towers. For the sake of completeness and
Fig. 2. The computational flowchart for PSO and KHO methods.
because these techniques can be directly applied to solar cavity receiver
geometry optimization, these papers are also included in the review.
Farahmand and Payan (2012), employed a PSO algorithm to optimize efficient than the mGA and PSO algorithms.
the geometrical shape of radiative enclosures having gray, diffuse sur­ Other attempts worth mentioning here that have been carried out for
faces. The aim was to optimize the shape of the cavity in order to satisfy the geometric optimization of radiant enclosures were performed by the
the desired temperature and heat flux distribution over the cavity’s team of Daun et al. (2003). In these two studies though, stochastic
surface. Their results showed that the fitness was improved when using gradient-based optimization methods were used, paving the way for
PSO, compared to the use of micro Genetic Algorithm (mGA) that was more advanced techniques (i.e. PSO) when these methods either fail or
performed earlier by Sarvari (2007). In the latter study, the author become difficult to apply.
solved the optimization problem discussed in Farahmand and Payan
(2012) by using a mGA, which allowed sizes of the population that were 3.1.2. External Receivers
very small, while the population being independent on the initial value The simplest form of external receiver is the cylindrical receiver,
and on the derivatives of the objective function. They concluded that the which represents the state-of-the-art in solar tower systems. Most of the
performance of the PSO was better than the performance of the mGA in optimization efforts on these receivers have focused on finding the
terms of convergence speed, mainly because the PSO algorithm has no optimal heat flux distribution on the receiver’s surface by employing the
complicated evolutionary operators, such as crossovers or mutations. Qi optimization of heliostat aiming strategies (i.e. Collado and Guallar
et al. (2016) considered PSO algorithms to perform an inverse geometry (2018)), as will be discussed in 3.2.2. Cylindrical receivers have convex
design of radiant enclosures, attempting to achieve a uniform distribu­ shapes which greatly simplify the radiative heat transfer simulations.
tion of radiative heat flux on the designed surface. Apart from the For external receivers with partly concave geometries, the same diffi­
standard PSO algorithm, authors also conducted a comparison between culties as for cavity receivers arise. To overcome this issue, Asselineau
a stochastic PSO (SPSO) and a differential evolution PSO (DEPSO), et al. (2016, 2017), presented receiver geometry optimization results for
which showed that the SPSO achieved the best performance in terms of a surround field configuration, using techniques originating from Sim­
computational time. Fig. 2a shows an overview of the main procedure ulation–Optimization Algorithms (SOA) that are dedicated to solving
that the authors used. A combination of MCRT simulations and a GA was ”black-box” problems. The authors used an algorithm that utilizes a
also employed in the study of Darvishvand et al. (2016) in an attempt of machine learning approach to find the best receiver candidate that
optimizing the uniformity of thermal conditions within three dimen­ would perform better in relation to the existing candidates, taking into
sional radiant enclosures. The results showed that the technique was account the known information. This approach was capable of identi­
capable to accomplish an optimal thermal condition on the surfaces of fying receiver candidates with strong potential, while the analysis of the
the enclosure. results indicated that improved receiver designs could be obtained by
The Krill Herd Optimization (KHO) approach is another bio-inspired careful management of the receiver’s flux distribution with respect to
algorithm that mimics the herding behavior of each individual krill for the receiver’s geometry. Fig. 3 shows the best receiver geometry can­
finding the minimum distance from food, thus is being used for opti­ didates as obtained in Asselineau et al. (2017) in terms of second law
mizing relevant problems. The KHO algorithm was employed by Sun efficiency, minimal average temperature, minimal area and best overall
et al. (2016) for the inverse geometry design of two-dimensional com­ ranking respectively.
plex radiative enclosure (Fig. 2b). In their study the authors used the Luo et al. (2017) studied the optimization of a heated plate receiver
discrete ordinate method with a body-fitted coordinate system along geometry considering both dimple/protrusion depth and delta-winglet
with different kinds of KHO algorithms to optimize the shape. They vortex generators on the surface of the receiver. The optimization ob­
concluded that KHO algorithms can be successfully employed for solving jectives involved were the thermal performance of the receiver and the
inverse geometry design problems and showed that they are more outlet dimensionless turbulent kinetic energy. The geometrical

504
K. Milidonis et al. Solar Energy 224 (2021) 500–515

optimization variables, while a combination of Computational Fluid

Dynamics (CFD) and a NSGA was chosen for the optimization sequence
(Fig. 4).

3.1.3. Volumetric receivers

Volumetric receivers have the potential to reach very high temper­
atures and hence achieve higher system efficiency. This advantage
comes mostly from the use of high-temperature metallic or ceramic
based absorbers and the use of air as a HTF. The application of AI
techniques for the optimization of the geometry of volumetric receivers
is very limited. The only available studies found in the literature are the
ones conducted by the team of Du et al. (2017, 2018). In Du et al. (2017),
the authors used a GA for optimizing the distribution of the porosity of a
volumetric solar receiver. The optimization process led to a much higher
depth penetration of the solar radiation within the receiver. In Du et al.
(2018), the combination of a GA coupled with a classical optimization
method and a heat transfer analysis regarding the porosity of a volu­
metric receiver was employed for maximizing the receiver’s thermal
efficiency while minimizing the flow resistance through the receiver. A
volume averaging simulation method based on the local thermal non-
equilibrium model was used for the evaluation of the heat transfer and
fluid flow characteristics within the receiver. The performance of the
employed method in optimizing the receiver’s thermal efficiency was
shown.

3.2. Heliostat Field Optimization

3.2.1. Heliostat Field Layout – Spatial distribution of heliostats

The heliostat field represents 30–40 % of the cost of a solar tower
plant (Kolb et al., 2007; IRENA, 2012; Pfahl, 2014), while simulta­
neously incurring roughly 40 % of the losses in the energy conversion
chain. Therefore, optimization of the field layout is critical. The objec­
tive of the heliostat field design is to maximize the annual optical field
efficiency, with two objective functions being predominantly used in the
literature: the annual weighted and the annual unweighted heliostat
field optical efficiencies.
Fig. 3. Best receiver geometry candidates as obtained from Asselineau et al. The annual weighted heliostat field optical efficiency is defined as:
(2017) in terms of: (a) Work efficiency, (b) Minimal area and (c) Best .over­
365 ∫
∑
all ranking. sunset
η
I(t) (t) dt
sunrise
ηyear,w = d=1
(1)
∑ ∫ sunset
3 65
characteristics of the winglet and the dimples were selected as I(t)dt
sunrise
d=1

where d refers to the day of the year and I(t) is the instantaneous solar
radiation.
The annual unweighted heliostat field optical efficiency is defined as:
365 ∫
∑ sunset
sunrise
η(t) dt
ηyear = d=1365 (2)
∑ ∫ sunset
sunrise
dt
d=1

Additional works considered the extension of the annual heliostat field

efficiency optimization so that it includes a joint optimization of receiver
parameters, such as tower height and receiver aperture, heliostat aiming
points, and field or heliostat costs in order to optimize the cost of the
thermal energy delivered. The optimization of the heliostat field layout
framed in this way is a highly challenging problem, involving non-
convex constraints, a non-smooth multi-modal objective function that
does not have a closed form, and a large number of variables (typically
thousands of heliostats in a field) (Carrizosa et al., 2017). Four main
methods for field layout optimization have been proposed in the liter­
ature (Lutchman et al., 2014): the field growth method, the pattern-
based method, the free variable method, and hybrid methods based on
a combination of any two of the previous methods.
Fig. 4. Optimization procedure using CFD and NSGA (Luo et al. (2017)). Field growth methods The field growth method is initiated by

505
K. Milidonis et al. Solar Energy 224 (2021) 500–515

considering the maximum potential energy that every location within

the field can deliver. The first heliostat is then positioned in the location
where it can deliver the most energy. The field is then re-evaluated,
taking into account the shading caused by the first heliostat, until the
second best location is found and a heliostat is placed there. When the
re-arrangement is completed, the whole field is evaluated to determine if
system requirements are met: if the conditions/requirements are met,
the process is terminated, otherwise it is repeated. Although this method
gives good results, it is computationally expensive, with the computa­
tional cost increasing exponentially with each placement. Implementa­
tions of this method are considered in the literature i.e. (Sanchez and
Romero, 2006; Yao et al., 2015), with traditional gradient-based opti­
mization methods being used to determine optimal placement location
in each step.
Pattern-based methods In pattern-based methods, a pattern is used to
place the heliostats on the field, with the spacing parameters being
considered as the variables to be optimized. The pioneering work of
Lipps and Vant-Hull (1978) considered four patterns: radial and
orthogonal cornfields and staggers, concluding that the radially-
staggered pattern results in the highest annual efficiency. This work
defined the patterns considered until Noone et al. (2012) introduced a
biomimetic spiral pattern, having two shape parameters to be optimized.
By limiting the degrees of freedom, since the locations where heliostats
can be positioned are defined by the pattern chosen, the complexity of
the problem is reduced, making this one of the most popular methods in
the literature. However, limitations of this method exist, since taking
terrain variation or time asymmetric weather conditions into consider­
ation is more challenging with this approach (Romero et al., 2002).
Several authors have worked on optimizing the various pattern
spacing parameters through the use of conventional gradient-based
methods (i.e. (Wei et al., 2010; Noone et al., 2012; Collado and Gual­
lar, 2012; Collado and Guallar, 2013; Besarati and Goswami, 2014;
Lutchman et al., 2014)). Recently, AI methods have been used to opti­
mize the pattern-based layouts. One study performed by Pitz-Paal et al.
(2011) considered a field for thermochemical processes, where the
objective function included field performance, secondary reflector per­
formance and receiver parameters. The field was optimized by
combining a GA and the Nelder-Mead algorithm. Talebizadeh et al.
(2014) also used GA when considering receiver parameters in addition
to field performance parameters in the optimization.
Free-variable methods The free variable method considers a random
initial population and the location of each heliostat is updated in each
iteration without following a predefined pattern. Even though the large
number of variables and the non-continuous nature of the evaluation
function makes this method computationally challenging, AI algorithms
are considered to be an elegant choice to solve these problems.
PSO methods (Farges et al., 2013), EA (Richter et al., 2016b), GA
(Lazardjani et al., 2016, ?, ?)), iterative artificial bee colony (ABC) al­
gorithms (Arrif et al., 2018), have all been applied to the free variable
method for field layouts. Recent developments to improve these
Fig. 5. Comparison of optimization algorithms on PS10 heliostat field (Richter
methods involve the implementation of a new crossover method to
et al., 2016b).
accelerate the convergence of an advanced genetic algorithm as shown
by Wang et al. (2018), and the parallelization of the optimization codes
performed by Cruz et al. (2018). Fig. 5 provides a comparison of PS10 restricted fashion so that an optimal spacing is found. This problem
heliostat layouts achieved using different optimization techniques as has been approached by conventional gradient-based optimization
obtained by Richter et al. (2016b). methods (i.e. (Buck, 2012)), but also through application of AI. Atif and
Hybrid methods In the fourth category of combined methods, the field Al-Sulaiman (2015) applied an EA, named differential evolution,
growth method is coupled with a free variable method in the work applied to a radial-staggered initial field layout, whereas others used
performed by Carrizosa et al. (2015a), where a greedy-heuristic algo­ either a PSO (Izygon et al., 2018; Kiwan and Khammash, 2018), or a
rithm considering the best field location is used to generate initial combination of PSO with GA (Li et al., 2018; Li et al., 2017). In the latter
populations. In an interesting extension, of the latter work, multiple studies, the hybrid PSO-GA algorithm was used in order to combine the
sized heliostats were also considered to be effective on using the avail­ merits of the different optimization algorithms but at the same time
able land (Carrizosa et al., 2017). avoid their shortcomings. Fig. 6 illustrates the flow chart of the hybrid
The pattern-based method is most readily coupled with the free- PSO-GA employed by the authors, while pbest and gbest refer to the
variable method, using the former to establish a good initial field heliostat previous best position and group previous best position
layout and subsequently allowing the heliostats to move in a non- respectively.

506
K. Milidonis et al. Solar Energy 224 (2021) 500–515

Fig. 6. Chart flow of a hybrid PSO-GA technique employed by (Li et al., 2018).

An interesting combination of gradient-based (Rosen algorithm) and optimization objectives set.

simulated annealing algorithm has been reported by Zhang et al. (2016) Finally, for analyzing the optical performance and obtaining optimal
in order to accelerate the convergence process of optimizing and heliostat fields for tower systems on hilly, mountainous terrains, Kiwan
generating various types of heliostat fields by accomplishing compari­ and Al Hamad (2018) employed PSO and biomimetic spiral distribution
sons between them. Among their results, the authors reported that a pattern algorithms. The overall optimization method was considering
combination of biomimetic spiral and staggered heliostat layouts offers the effects of multiple variables, including the location and tilt angle of
a potential to increase the efficiency of the heliostat field. A schematic the receive, the angle of the terrain etc. Among others, the authors found
diagram of the Rosen-SA algorithm that was employed is shown in that an optimum hillside angle exists for which the land area required is
Fig. 7. Moreover, Deng et al. (2020) explored the capabilities of a set of minimized. Based on analysis of site slope effect, Lee and Lee (2019)
differential evolution algorithms for optimizing a rose pattern-based employed a surrogate model and GA for determining the most efficient
heliostat field in which the heliostats are radially staggered within six heliostat field type for each latitude location while optimizing the he­
sectors, with each sector being optimized separately. Their results liostat field layout on the optimal site. The results showed that the se­
showed that the strategy of the dynamic speciation-based mutation lection of the most efficient site has a direct impact on the optical
differential evolution (DSM-DE) algorithm achieved a higher annual efficiency, and the percentage increment of optical efficiency is greater
weighted efficiency. In addition, Schöttl et al. (2019) used an EA to than can be achieved by only optimizing the heliostat field only.
optimize the vertices of a polygon shaped heliostat field by taking into
account the optimal trade-off between land usage and annual optical 3.2.2. Heliostat Positioning/Alignment - Aiming
efficiency of the field. The authors illustrated that the EA was able to Accurate control of heliostats is one of the most important tasks in
design a coherent and smooth heliostat field that meets all the solar tower systems. Examples of control tasks include the selection of
aiming points on receiver and the decision on activation strategy so as to
avoid dangerous radiation peaks and severe temperature gradients on
the receiver surface. In case of multi-receiver or multi-tower systems, the
control of heliostats is particularly complex, because the minimization
of cosine losses can increase the shading and blocking losses substan­
tially. For these tasks, the use of AI techniques has been found to offer
unique pathways to address relevant challenges.
Heliostat actuators design and controls For estimating the accuracy of
the orientation of heliostats in a heliostat field, Zeghoudi and Chermitti
(2014) employed a combination of an open and a closed loop control
system using an ANN algorithm. The ANN was trained using the Lev­
enberg - Marquardt (LM) ((Levenberg, 1944; Marquardt, 1963)) back
propagation algorithm with several parameters influencing the path of
each heliostat. These parameters included the date and position of the
sun, the location of the plant, the elevation and azimuth time as well as
Fig. 7. The schematic diagram of the Rosen-SA algorithm employed by Zhang the tower’s height and heliostat’s slope and size (Fig. 8). They reported
et al. (2016). The Rosen algorithm is initially used for optimizing in a few steps excellent predictions of the movement of the heliostats, concluding that
the initial random solution for obtaining a sensible solution, and, later the SA the method is suitable for estimating, controlling and monitoring of the
algorithm operates a local random search.

507
K. Milidonis et al. Solar Energy 224 (2021) 500–515

performance while reducing the overshoots. Furthermore, it was found

that the F-PI controller is insensitive to noise, thus making it a proper
choice for controlling heliostats. An example of the block diagram that
the authors used is shown in Fig. 9. The use of an adaptive F-PI controller
was also described in studies performed by the team of Zeghoudi et al.
(2016, 2015), in which the performance of the F-PI controller is
compared to the performance of the conventional Proportional-Integral
(PI) controller. During their tests, two different approaches for the fuzzy
controller were taken into account. Their results showed that it is
possible to further increase the system performance by adapting classic
controller parameters and changes in the rules of the FL. Moreover,
Huang et al. (2009) presented a solar tracking control system whose
fuzzy-based tracking controller was designed and implemented on a
Fig. 8. Neural network model for estimating heliostat azimuth and elevation Field Programmable Gate Array (FPGA) platform. Based on their
angles (adapted from Zeghoudi and Chermitti (2014). experimental work, the authors concluded that the proposed system
architecture achieves higher generated power efficiency. For high pre­
heliostat field of a solar tower system. Gonzalez-Tokman et al. (2014) cision heliostat tracking, Freeman et al. (2014) presented a novel two-
developed an ANN to control a heliostat field. The authors made use of level control architecture using a hybrid ANN and Fuzzy Inference En­
advanced techniques such as Extended Kalman Filter (EKF), PSO, gine (ANFIS) coupled with NREL’s Solar Position Algorithm (SPA). This
Recursive High Order Neural Networks (RHONN) and Sliding Modes approach consists of a hierarchical control architecture for which a
Control (SMC). A control law based on SMC and RHONN was used to model based controller occupies the supervisory level. As an objective,
track the trajectory of the sun. A RHONN, trained using EKF was used to the supervisory control had to aim the heliostats to the required position
model the motors of the heliostats. Finally, a PSO algorithm was used to with the proper accuracy in order to reflect the direct normal irradiation
obtain the initial parameters for the EKF. Their simulations showed that (DNI) from the sun to the central receiver. The conclusion drawn by
this approach is powerful for increasing the efficiency of the heliostat applying the approach is that the ANFIS system offers reduced
field because of its ability to address disturbances caused by external complexity and allows for the quick simulation of various heliostat ge­
parameters such as the wind, and its ability to absorb unmodeled dy­ ometries. In addition, it seems that one of the benefits of using ANFIS
namics of the system. Furthermore, Guangyu and Zhongkun (2017) methods is that highly accurate actuator results can be obtained without
proposed a closed-loop feedback control method based on machine solving complex kinematic equations. These advantages translated into
vision methods for correcting the aiming of a heliostat focusing on a test cost reductions with limited calculation complexity and computational
target. By employing a Back Propagation Neural Network (BPNN), the time. A FL based stepper motor drive was employed by Jirasuwankul
control system was able to accurately determine the characteristics of and Manop (2017) for a lab scale heliostat positioning, in which the FL
the heliostat flux distribution on the target by mapping the azimuth and was also acting as a tracking supervisor and was responsible for the
elevation angles. Thus providing direct feedback control of the heliostat multi-stepping rate adjustment of the stepper motors. In addition, the
positioning errors. closed-loop tracking system was using a video streaming device in
Recently, Carballo et al. (2019) presented a new approach for solar conjunction with an image analyzer. The application of the technique
tracking based on deep learning and computer vision. This new showed that minimization of tracking errors can be obtained.
approach relies on deep Convolution Neural Network (dCNN) to localize Heliostat field aiming Optimizing the aiming points of the heliostats is
and detect the objects. The basic idea behind this method was to train a challenging mathematical task. To reduce the complexity and the
the dCNN with multiple heliostat position images, taken while the he­ computational effort required to assign the aim point of every single
liostat was being operated by a traditional control system. Then, all heliostat in a heliostat field, respecting peak flux and thermal input
images are analyzed in sequence by the CNN object detector in order to constraints while maximising optical efficiency, numerous parameteri­
acquire all the classes which would allow the development of the sun zation methods have been used. Among them, a popular approach is to
tracking task in an appropriate manner, while searching for key pa­ transform the continuous problem into a combinatorial problem,
rameters originating from other heliostats (shadowing or blocking) or allowing the application of faster algorithms. Typically, the receiver
other parameters such as clouding effects. It was concluded that the target is subdivided into discrete aim-points and heliostats are optimally
proposed approach is credible and functional, albeit with significant ”assigned” to these aim-points. Based on this approach, Belhomme et al.
room for improvement. The same authors, for evaluating their advance (2014) adapted an ACO metaheuristic approach to optimize the problem
control methods described above, developed a mock-up solar tower of heliostat aiming of a solar tower system, considering the performance
system (Carballo et al., 2020) which was composed of model 3D printed of the receiver as a function of the distribution of the flux density. The
structures of the heliostat and tower, and the combination of low cost Ant Colony Optimization (ACO) method is a bio-inspired metaheuristic
hardware and open source software. The team of Carballo et al. (2019) optimization approach originating from the behavior of real ants, and
also employed the training and tuning capabilities of the open source more specifically their ability to find the optimal, i.e. shortest path,
machine learning framework called ”Tensorflow”, for investigating a between their nest and food. Artificial ACO algorithms are thus usually
new approach in sun tracking. A small mock-up heliostat was used for employed for solving and optimizing shortest path problems. Among
training the NN involved. The authors concluded with promising results
in terms of the ability of the NN to properly track the apparent trajectory
of the sun without further information.
Fuzzy logic has been used to enhance sun tracking by heliostats.
Ardehali and Emam (2011) developed a fuzzy based controller to
improve the azimuth and elevation tracking behaviour of heliostats.
Two types of controllers were considered in their study: a fuzzy
Proportional-Integral-Derivative (F-PID) and a fuzzy Proportional-
Integral (F-PI) controller. They concluded that for both controllers the Fig. 9. Block diagram for controlling the azimuth and elevation axes (adapted
application of FL based on tuning the gain coefficients can lead to better from Ardehali and Emam (2011). (For interpretation of the references to colour
in this figure legend, the reader is referred to the web version of this article.)

508
K. Milidonis et al. Solar Energy 224 (2021) 500–515

others, the authors pointed out that this technique facilitates the path developed could perform all the necessary calculations and activities for
towards real-time optimization during the operation of the plant, controlling the heliostat orientation (with a mean square error of less
contributing to a significant efficiency increment. Moreover, the capa­ than 0.1 degree) by the use of only a single low cost microcontroller. In
bility of the ACO metaheuristic on adapting itself to the dynamically another study Salgado-Plasencia et al. (2020), the same team used a
changing variables and conditions is highlighted. The optimal heliostat digital signal processor microcontoller-based FL controller also for
strategy for achieving optimal distribution of heat flux on the receiver controlling the orientation of heliostats. In this case, experimental re­
was investigated by Besarati et al. (2014). In their study, a new opti­ sults showed the superiority of the method in terms of flexibility and
mization algorithm operating on principles of GA (and modified to the performance over the use of traditional PID controllers. Zhu and Ni
physics of the problem) was introduced. The main objective of this al­ (2019) used a GA to optimize the aiming strategy of heliostats for
gorithm was to minimize the standard deviation of the flux density achieving steady solar fluxes according to the power requirements of the
distribution on the receiver surface. It was shown that the algorithm is system, while the calculation regarding the solar flux was provided by
very effective for different numbers of aiming points, concluding that for an MCRT-based software. Their results illustrated the superiority of the
better performance the number of the aiming point should be increased ANN against the conventional methods for optimizing the aiming
as much as possible. In an attempt to homogenize the solar flux distri­ strategy of heliostats. Finally, Cruz et al. (2019) also employed a GA to
bution inside a cavity receiver, Wang et al. (2017) optimized an aiming optimize the distribution of the aiming points of a heliostat field in order
strategy through the use of a non-dominated GA. The distribution of the to achieve an optimal combination of a desired flux distribution on the
discrete aiming points on the cavity aperture and the allocation of the receiver, while eliminating possible flux peaks which could become
aiming points for each heliostat were optimized simultaneously. The dangerous in terms of the structural integrity of the receiver.
solar flux was calculated through MCRT, and the fast elitist non- Table 3 summarises the applications of AI techniques to heliostat
dominated sorting genetic algorithm (NSGA-II) (Deb et al., 2000) was field applications. It can be seen that FL is the most widely used algo­
applied to optimize the aiming strategy. In addition to the successful rithm for this purpose, while ANN and GA methodologies are also
homogenization of the solar flux on the receiver surface, the proposed frequently used.
method provided a trade-off between the non-uniformity of the flux
distribution and the optical losses in the form of Pareto optimal front
3.3. Multi-Tower Receivers
curves.
Cruz et al. (2018) employed a two stage optimization method to
The multi-tower receiver concept is a system architecture for large
solve the problem of automatic heliostat aiming. The proposed method
solar tower applications whose main characteristic is the increment of
decides which heliostats should be activated and where they should aim
the heliostat and land utilization efficiencies. This is achieved by
in order to replicate any given reference flux distribution on the
reducing the attenuation and spillage losses that would normally grow
receiver. During the first stage, a GA was employed in order to handle
when increasing the distance of the heliostats from the central receiver.
the initial heliostat selection, activation and assignation of acceptable
The technology associated with the multi-tower systems is relatively
aim point, whereas a gradient descent (GD) method was adopted during
new, hence the use of AI techniques for the optimization of such plants is
the second stage as a resolution level to further improve the quality of
still in early stages and in the literature, only a couple of articles can be
the solution obtained by the first stage. The latter stage was imple­
found. Multi-tower configurations add complexity on top of the existing
mented through fine adjustment of the active heliostats. After applying
difficulties through added design parameters and constraints, increasing
the method on a virtual field, the authors concluded that the method is
the necessity of using advanced optimization techniques.
innovative in various aspects, and that it is a powerful tool to define
A specific problem of multi-tower systems is that heliostats aiming at
regular operational states and to assist human operators. An early study
different receivers can drastically shade and block each other. To
by Garcia-Martin et al. (1999) presented the development and imple­
minimize the effect of this problem, Piroozmand and Boroushaki (2016)
mentation of a knowledge-based heuristic for a heliostat control system,
proposed a method for the optimal design of a multi-tower heliostat
focusing on the optimization of the temperature distribution within a
field. The mathematical procedure that they employed for optimizing
volumetric receiver. In order to avoid the use of traditional automatic
the shape parameters related to the pattern of the heliostat field as well
control systems, the authors focused on developing a heuristic control
as for the distance separating the towers was based on a PSO evolu­
strategy approach in which the performance of a skilled operator during
tionary algorithm. The authors applied the method on a case study of a
the tests could be reproduced. Their approach was applied to the volu­
two-tower spiral pattern field and their results revealed an improvement
metric receiver of the CESA-1 plant (Plataforma Solar de Almeria) and it
in the annual efficiency and optical performance of the field by letting
was proven that the strategy allowed the objectives set to be fulfilled
the heliostats make decisions individually for selecting the proper
while avoiding the continuous involvement of the plant operator.
receiver to aim at different times of the days. Fig. 10 shows the heliostat
Monterreal and Jimenez (2003) used a GA in order to find the
behavior at different time periods during the day by using the above
optimal parameterization for heliostat optics. To achieve their goal, the
method. Carrizosa et al. (2015b) developed a new approach for opti­
authors compared real and simulated heliostat flux distributions on the
mizing the design of solar tower systems in the presence of multiple
target/receiver, after which the GA explored the optimization domain
receivers by simultaneously optimizing the receivers and the heliostat
with a certain confidence and guarantee. It was shown that an approx­
field. More specifically, the optimal arrangement of the heliostats was
imation to the real model of the heliostat optics can be obtained, thus
obtained through a pattern-free greedy-based location method, whereas
enabling reliable prediction of irradiance distribution on the solar
the aiming regions along with the number of heliostats were obtained
receiver surface. Soylemez and Aldulaimi (2016) employed parallel
from the proposed procedure. The authors tested their strategy on a
computing architectures and ANN to enhance the optical performance of
three receiver configuration, showing that the resulting overall plant
the heliostat field under shading and blocking. Their aim was twofold:
architecture had lower LCOE value compared to the initial
reduce the implementation time and discard unnecessary computations
while maintaining high accuracy. The investigation of the proposed
Table 3
system showed higher efficiency in the process of heliostat field design
Summary of algorithms used for heliostat positioning.
optimization for a lower computation time. For controlling the orien­
tation of heliostats, the team of Salgado-Plasencia et al. (2019) designed Algorithm Count Algorithm Count

and constructed a Supervisory Control and Data Acquisition (SCADA) ANN 4 FL 8

control system in combination with an orientation control implemented NN + PSO 1 ANFIS 1
GA 5 ACO 1
with fuzzy a logic controller (FLC). It was shown that the system

509
K. Milidonis et al. Solar Energy 224 (2021) 500–515

proposed method offers the possibility for different operation modes of

the plant, due to a mid-term storage system that balances the continu­
ously changing DNI to a step wise changing power output. Cavallaro
(2010) employed a fuzzy TOPSIS (Technique for Order Performance by
Similarity to Ideal Solution) approach for the comparison of different
HTFs in order to investigate the feasibility of utilizing a molten salt as a
storage medium. The authors performed an empirical optimization case
study in which they compared different heat transfer fluids (HTFs) based
on their properties, costs and benefits. They concluded that the standard
molten salt offers the potential to reduce electricity production cost and
to increase the energy performance in an environmentally friendly way.

3.5. Maintenance and Diagnosis

Maintenance and diagnosis of solar tower systems are very important

activities in order for the plant to achieve the required reliability,
availability and safety. Major issues associated with these activities
usually involve predicting the possible defects in different components
of the plant, monitoring of the conditions, soiling levels of the reflecting
mirrors etc. The applicability of AI in this area is currently limited. Only
a handful of papers can be found in the literature.
Very recently, Ashley et al. (2019) introduced a heuristic approach in
order to optimize the heliostat cleaning schedule of a solar tower plant
due to soiling on their surfaces. The objective was to optimize the optical
efficiency, and thus, maximize the energy generated by the plant over
the schedule duration. The heuristics were used to produce routing
Fig. 10. Heliostats behavior at different time periods during a day, by using the
refinement options to improve the routes across the schedule. The au­
proposed method. Blue heliostats are aiming at the left receiver and red he­
liostats are aiming at the right receiver (Piroozmand and Boroushaki, 2016).
thors found the technique provided an increase of nearly 5 % in total
(For interpretation of the references to colour in this figure legend, the reader is energy compared to a traditional cleaning schedule, which just followed
referred to the web version of this article.) the rows of the heliostats in the plant. Conceicao et al. (2018) measured
mirror soiling as a function of environmental parameters and developed
configuration. models to predict the soiling based both on multiple linear regression
A study worth mentioning here is the one performed recently by and on ANN. Bonanos et al. (2019) performed a similar analysis,
Kiwan and Latif Khammash (2019) who instead of considering multiple correlating mirror soiling to environmental parameters using a feed-
tower receivers, they considered the optimization of a novel idea of forward back-propagation neural network. They also used their model
incorporating two receivers mounted on a single tower at different to predict the mirror cleaning frequency required by the plant operator
heights. For investigating this idea the authors used a PSO algorithm for to maintain the mean field reflectivity above a particular value, thus
optimizing the field’s shape factors. As a test case for the idea, the au­ helping in estimating the field operational and maintenance costs.
thors employed a 50 MWth reference tower plant for which it was shown Taking into advantage the benefits of forecasting of the soiling rate,
that the double-receiver system yielded a weighted optical efficiency of Terhag et al. (2019) employed an ANN algorithm to optimize the
almost 67.5% in comparison to the 67.14% that was yielded for the cleaning strategy of the heliostat mirrors for the PSA field. The results
single receiver configuration. This corresponded to a land saving of showed an increment of 1.33% in terms of the project’s profit if the
nearly 11000m2 and to a reduced number of heliostats needed to obtain algorithm is fed with a single day soiling-rate forecast. In terms of he­
the same power. liostat field cleaning strategies, another work worth mentioning is the
one conducted by. Picotti et al. (2020) who employed a Mixed Integer
Linear Programming (MILP) in an attempt to optimize the cleaning
3.4. Thermal Energy Storage schedule of the field. Their results showed that the algorithm was able to
provide a reduction of the total cleaning cost.
The Thermal Energy Storage (TES) system is the subsystem of a CST In a study performed by Pai et al. (2013), a Cerebellar Model
system that stores as thermal energy a fraction of the solar energy Arithmetic Computer (CMAC) NN was proposed for automatic diagnosis
delivered to the receiver(s) by the solar concentrators (heliostats, of faults within the heliostat field. For the diagnosis, several parameters
parabolic troughs, Linear Fresnel, parabolic dishes, etc.) of the CST related to the transmission system of the heliostats’ were taken into
system, to be used when it makes more technical or economic sense. The account, such as vibrations, temperature characteristics and rotational
most common thermal storage medium employed in CST systems and speed. Their experimental results suggested that the proposed system
particularly in solar towers is the so called ”solar salt” a mixture of has a short learning time, a high classification performance and a good
NaNO3 and KNO3 salts, but other mediums (liquids or solids) are also robustness to noise in the input signal. They also reported that the CMAC
used such as water, sand, rocks, concrete etc. Optimization goals for the neural network method they employed requires shorter training times
TES system include improving performance through efficient control than the Radial Basis Function (RBF) or the BPNN. Song et al. (2012)
loop, the overall optimization of the seasonal storage strategy, etc. coupled a machine vision-based diagnosis system with Charge Coupled
(Cirocco et al., 2014). However, the application of AI techniques to this Devices (CCD) for automatic heliostat fault detection and diagnosis of a
subject is still very limited. solar tower system. The CCD was used for monitoring the heliostat field
Schlipf et al. (2014) developed and demonstrated a feed forward and detecting faulty heliostats via processing of the images. Their
control system for which the optimization loop was based on a GA. The implementation of this algorithm on a heliostat field showed that the
system was applied on a direct steam generation CSP system that had system can be used to detect faulty heliostats in large heliostat fields
mid and long term storage capabilities. The authors concluded that the which is composed from hundreds to thousands of heliostats.

510
K. Milidonis et al. Solar Energy 224 (2021) 500–515

3.6. Optimization of hybrid solar tower and photovoltaic plants would yield a minimal LCOE.
Ramos and Ramos (2012) considered different algorithms, particu­
In comparison to the CST and photovoltaic (PV) standalone tech­ larly two local optimizers and a global one, to optimize a solar tower
nologies, it is acknowledged that integration of the two technologies system focusing on the design of major components of the plant, namely
may unleash huge potential in terms of the cost competitiveness, the heliostat field, the tower and the receiver. The plant was parame­
generated efficiency, quality, etc. For this subject, less than a handful of terized as a function of eleven design variables, whereas a GA algorithm
papers can be found in the literature so far. was chosen as the global algorithm to prevent the local optimization
For minimizing the levelized cost of electricity of a hybrid solar techniques from falling into local optima. This approach was reported to
tower system with PV, Liu et al. (2019) employed a GA-PSO algorithm. be robust and sufficiently fast to tackle the complexity of the proposed
The optimization was focused on the nominal power of the PV compo­ non-linear objective function. Furthermore, it was demonstrated that
nent, the overall system power output and the capacity of the thermal this algorithm can be used to evaluate the impact of the design param­
energy storage in two different dispatch operating modes, an indepen­ eters on the plant performance. Richter et al. (2016a) adopted a GA to
dent output and a constant output mode. In short, for the independent estimate the annual performance of a heliostat field as a function of
mode, the CST and PV components provide power independently of each topographical variables, tracking errors, sun’s position and DNI. The
other, while in constant output mode, the power output is kept constant proposed implementation allowed for heliostat field optimization within
by scheduling appropriately the outputs of the two sub-component. The few hours, mainly due to the reduction of CPU time needed to estimate
algorithm was able to estimate the operating performance of the hybrid the shading and blockage effects. In the work of Moukhtar et al. (2017),
solar tower and photovoltaic (CST-PV) plant, indicating that the per­ an ANN is used for adjusting the HTF mass flow rate for any solar
formance of the constant output mode is steadier than the independent thermal power input to the receiver, so that the temperature of the HTF
mode. For solving a similar problem, Wang et al. (2019) used a multi- at the receiver output is adjusted at the estimated value. The ANN was
objective PSO algorithm considering at the same time the cost and capable of maintaining the receiver temperature output at the desired
performance of the hybrid plant, while the two sub-components of the value during the year, thus being able to estimate the annual perfor­
plant are working synergistically to meet the power demand. The opti­ mance of a solar tower system in a simple and flexible manner. Recently,
mization algorithm in this case was also fed with real experimental data a PSO algorithm was used by Farges et al. (2018) so that global optima
in terms of the plant’s site meteorological and solar resource data. The could be achieved for the design of a solar tower systems. The perfor­
results obtained illustrated the ability of the algorithm to provide a mance was evaluated in terms of two performance indicators: the yearly
reasonable design of the hybrid system. heliostat field optical efficiency and the thermal energy collected
annually by a solar tower system. The results showed that a significant
improvement was achieved for both performance indicators. Luo et al.
3.7. Overall Concentrated Solar Thermal tower plant optimization (2018) achieved an accurate design for the solar tower by combining a
Sobol sampling method along with a SA algorithm in order to accelerate
This section regroups additional works that adopted a more holistic the computational time needed for the SA algorithm to provide a global
approach to simultaneously optimize plant performance and minimize optimum over the highly multi-dimensional design space. The hybrid
the levelized cost of electricity (LCOE). technique leads to a substantial reduction of the number of optimization
Lopez-Martinez et al. (2002) considered artificial vision-based (AV) steps needed to achieve global optima compared to the classical SA al­
systems for locating and tracking the sun in order to estimate the he­ gorithm. Another study by Luo et al. (2019) also utilizes a global SA
liostat area that is shaded by passing clouds. They suggested that these algorithm along with a successive response surface methodology (SRSM)
techniques can help the master control system to reduce the number of to acquire the minimum possible Levelized Cost of Electricity (LCOE) of
heliostats aiming to the receiver while decreasing the incident radiation a solar tower system. Their results revealed the superiority of the pro­
so that a rupture of the receiver due to thermal stresses is prevented. posed optimization method compared to the actual model-based algo­
Artificial vision algorithms were used for the location and tracking of the rithm in terms of accuracy and computation run time.
sun, as well as the segmentation of clouds. A hybrid model for the Moukhtar et al. (2018) investigated the performance of a ANN-based
determination of economically optimal design parameters was proposed control scheme on predicting and simulating a solar tower system with
by Richter et al. (2011). In particular, they used a GA in order to obtain thermal energy storage. The MLP NN was chosen with the receiver
the global optimum of a solar plant configuration, whereas a Multi-layer thermal power and receiver inlet temperature used as input variables
Perceptron (MLP) (Fig. 11) was employed in order to describe the and the mass flow rate of the heat transfer fluid used as output variables.
thermodynamic processes of a solar plant, resulting to a substantial To train the NN, three different types of algorithms were employed: the
reduction of computational time compared to an approach without Quasi-Newton, the LM and the Scaled Conjugate Gradient (SCG). Out of
ANNs. It is worth mentioning that the authors considered up to twenty the three, the LM algorithm displayed its superiority in estimating the
design parameters (such as the solar collector field size and distance required performance irrespective of the variations in the direct solar
between the collector rows, the storage capacity, pressure and temper­ radiation and the temperature at the inlet of the receiver.
ature values at various points, the condenser size, etc) to achieve the For finding the optimum size of a solar tower plant, Khosravi et al.
optimization goal of finding a configuration of these parameters that (2020) used a combination of an ANFIS optimized method along with a
GA and a teaching–learning-based optimization (TLBO) algorithm for
simulating the overall behavior of the system. The proposed method was
fed with various input parameters, including the plant’s location and the
design point direct normal irradiation, and was trained to find the op­
timum figures in terms of design parameters which, among others
included the thermal energy storage hours and the solar multiple. For
calculating and analyzing the cooling performance of different solar
tower plant layouts, Boukelia et al. (2020) employed ANN models. The
configuration layouts concerned among others, the use of molten salt as
heat transfer medium in comparison with direct steam generation. The
authors concluded that results generated by the ANN were more accu­
Fig. 11. Topology example of a Multi-Layer Perceptron (MLP), adapted from rate than the ones generated by other regression models. López et al.
Richter et al. (2011). (2018) analyzed the water vapour influence on the transmissivity of the

511
K. Milidonis et al. Solar Energy 224 (2021) 500–515

solar irradiance in solar tower plants. For this, the authors employed an
ANN that was trained to calculate the atmospheric attenuation of the
solar radiation having as input variables various parameters such as the
height of the tower, the position of the sun, etc. Their results demon­
strated the strong capability of ANN’s to calculate the DNI attenuation
with very small errors.
Application of Artificial Intelligence techniques in solar tower sys­
tems have also found applications in terms of optimizing the hybrid
integration of CST with other systems. For example, Keshavarzzadeh
et al. (2020) used NSGA-II and NSGA-III genetic algorithms for tech­
noeconomic and environmental optimization of the integration of a
solar tower system to a fresh water production system. Moreover, Awan
et al. (2020) also employed ANN and fuzzy logic algorithms to perform a
technoeconomic comparison between a 100 MW solar tower plant and a
100 MW photovoltaic plant located on the same location.
Lastly, in a study performed by Wang et al. (2020), a deep learning
(DL) algorithm was employed in order to perform accurate forecasting of
the thermal power reaching the receiver of a concentrating solar tower
system. The proposed approach showed that the thermal power was
accurately predicted within a short term half hour forecast.

Fig. 12. Evolution of published articles about the use of AI methods in relation
4. Conclusions and Perspectives
to solar tower systems from 1978 to 2020.

The article presents a comprehensive review of the use of AI methods

It is our hope that this article can serve as a reference document for
for the design, optimization, control and O&M of the complete system
researchers, scholars and industry interested in using AI methods to
and sub-components of solar tower plants. It shows that a wide variety of
design, operate and improve solar tower plants.
AI methods are being employed in relation to the following aspects of a
solar tower plant:
Declaration of Competing Interest
• Receiver geometry optimization,including cavity, external and
The authors declare that they have no known competing financial
volumetric receivers,
interests or personal relationships that could have appeared to influence
• Heliostat field optimization, including heliostat field layout and he­
the work reported in this paper.
liostat positioning/alignment/aiming
• Multi-tower systems
Acknowledgments
• Thermal energy storage
• Maintenance and diagnosis
This work was partially supported by the European Union’s Horizon
• Hybrid solar tower and photo voltaic systems
2020 research and innovation programme within the context of the
• Overall solar tower system optimization
CySTEM ERA Chair project, under grant agreement No. 667942 and of
the INSHIP project, under agreement No. 731287.
More specifically, machine learning techniques for multi-objective
stochastic optimization and evolutionary algorithms are extensively
References
used for receiver design and optimization. Regarding heliostat field
layouts, the most widely employed optimization methods are genetic Ardehali, M.M., Emam, S.H., 2011. Development, design and experimental testing of
algorithms, while fuzzy logic is preferably employed for heliostat posi­ fuzzy-based controllers for a laboratory scale sun-tracking heliostat. Fuzzy
tioning/alignment control, followed by genetic algorithms and artificial Information and Engineering 3 (3), 247. https://doi.org/10.1007/s12543-011-0081-
x. ISSN 1616-8666.
neural networks. In terms of maintenance and diagnosis, the use of
T. Arrif, A. Benchabane, M. Germoui, B. Bezza, and A. Belaid. Optimisation of heliostat
artificial neural networks seems to be the preferred method to address field layout for solar power tower systems using iterative artificial bee colony
issues related to optimization of the heliostat field cleaning schedule and algorithm: a review and case study. International Journal of Ambient Energy, pages
1–16, sep 2018. doi:10.1080/01430750.2018.1525581.
fault diagnosis. Concerning overall solar tower system optimization,
Ashley, T., Carrizosa, E., Fernandez-Cara, E., 2019. Heliostat field cleaning scheduling
artificial neural networks and genetic algorithms are the state-of-the-art. for solar power tower plants: A heuristic approach. Appl. Energy 235, 653–660.
Lastly, the application of AI techniques on thermal energy storage, on https://doi.org/10.1016/j.apenergy.2018.11.004. ISSN 0306-2619.
multi-tower receiver systems and on the optimization of hybrid CST-PV Asselineau, C.A., Zapata, J., Pye, J., May 2015. Geometrical shape optimization of a
cavity receiver using coupled radiative and hydrodynamic modeling. Energy
tower systems is still very limited with only a few articles available in the Procedia 69, 279–288. https://doi.org/10.1016/j.egypro.2015.03.032.
literature. Asselineau, C.A., Zapata, J., Pye, J., 2015. Integration of monte-carlo ray tracing with a
AI methods are becoming very promising tools in the quest of opti­ stochastic optimisation method: application to the design of solar receiver geometry.
Opt. Express 23 (11), A437. https://doi.org/10.1364/oe.23.00a437.
mizing the design and operation of solar tower plants. Fig. 12 shows the C.A. Asselineau, C. Corsi, J. Coventry, and J. Pye. Multi-objective method for receiver
evolution of the number of the relevant articles published in the last 40 design optimization using n-dimensional stochastic pareto front detection. In 2016
years. It is evidently clear that during the last decade the curve follows ASME Power and Energy Conference, Charlotte, NC, 06 2016. doi:10.13140/
RG.2.1.2181.0168.
an exponential growth, meaning that researchers and engineers within Asselineau, C.A., Corsi, C., Coventry, J., Pye, J., 2017. Geometrical exploration of a flux-
the sector are turning to AI to address relevant challenges in the design, optimised sodium receiver through multi-objective optimisation. AIP Conf. Proc.
optimization and operation of solar tower systems. 1850 (1), 030002. https://doi.org/10.1063/1.4984345.
Atif, M., Al-Sulaiman, F.A., 2015. Optimization of heliostat field layout in solar central
There is still a lot of space to incorporate AI techniques to address the
receiver systems on annual basis using differential evolution algorithm. Energy
challenges that exist in all aspects of a solar tower system. In this regard, Convers. Manage. 95, 1–9. https://doi.org/10.1016/j.enconman.2015.01.089. ISSN
it is worth mentioning that recent and advanced machine learning 0196-8904.
techniques such as Random Forests or Gradient Boosting methods have
not yet found their way into the field of solar tower systems.

512
K. Milidonis et al. Solar Energy 224 (2021) 500–515

Awan, Ahmed Bilal, Zubair, Muhammad, Chandra Mouli, Kotturu V.V., 2020. Design, Cruz, N.C., Salhi, S., Redondo, J.L., Alvarez, J.D., Berenguel, M., Ortigosa, P.M., 2018.
optimization and performance comparison of solar tower and photovoltaic power Design of a parallel genetic algorithm for continuous and pattern-free heliostat field
plants. Energy 199, 117450. https://doi.org/10.1016/j.energy.2020.117450. optimization. The Journal of Supercomputing 75 (3), 1268–1283. https://doi.org/
Behrang, M.A., Assareh, E., Ghanbarzadeh, A., Noghrehabadi, A.R., 2010. The potential 10.1007/s11227-018-2404-8.
of different artificial neural network (ann) techniques in daily global solar radiation J. Cumpston and J. Pye. Exergoeconomic optimisation of steam networks connecting
modeling based on meteorological data. Sol. Energy 84 (8), 1468–1480. https://doi. solar-thermal dish arrays. Solar Energy, 119: 383–398, sep 2015. doi:10.1016/j.
org/10.1016/j.solener.2010.05.009. solener.2015.06.015.
B. Belhomme, R. Pitz-Paal P., and Schwarzbozl. Optimization of heliostat aim point Darvishvand, Leila, Kowsary, Farshad, Jafari, Pantea Hadi, 2016. Optimization of 3-
selection for central receiver systems based on the ant colony optimization d radiant enclosures with the objective of uniform thermal conditions on 3-d design
metaheuristic. Journal of Solar Energy Engineering, 136, 02 2014. doi:10.1115/ bodies. Heat Transfer Eng. 37 (1), 1–12. https://doi.org/10.1080/
1.4024738. 01457632.2015.1025002.
R. Belu. Artificial intelligence techniques for solar energy and photovoltaic applications. Daun, K.J., Howell, J.R., Morton, D.P., 2003. Geometric optimization of radiative
In Robotics: Concepts, Methodologies, Tools, and Applications, pages 1662–1720. enclosures through nonlinear programming. Numerical Heat Transfer, Part B:
IGI Global, 2014. doi:10.4018/978-1-4666-4607-0.ch081. Fundamentals 43 (3), 203–219. https://doi.org/10.1080/713836205.
Besarati, S.M., Goswami, D.Y., 2014. A computationally efficient method for the design Daun, K.J., Morton, D.P., Howell, J.R., 2003. Geometric optimization of radiant
of the heliostat field for solar power tower plant. Renewable Energy 69, 226–232. enclosures containing specular surfaces. J. Heat Transfer 125 (5), 845–851. https://
https://doi.org/10.1016/j.renene.2014.03.043. ISSN 0960-1481. doi.org/10.1115/1.1599369.
Besarati, S.M., Goswami, D.Y., Stefanakos, E.K., 2014. Optimal heliostat aiming strategy Deb, K., Agrawal, S., Pratap, A., Meyarivan, T., 2000. A fast elitist non-dominated sorting
for uniform distribution of heat flux on the receiver of a solar power tower plant. genetic algorithm for multi-objective optimization: Nsga-ii. In: International
Energy Convers. Manage. 84, 234–243. https://doi.org/10.1016/j. Conference on Parallel Problem Solving from Nature, pp. 849–858. https://doi.org/
enconman.2014.04.030. ISSN 0196-8904. 10.1007/3-540-45356-3-83.
Boito, P., Grena, R., 2016. Optimization of the geometry of fresnel linear collectors. Sol. Deng, L., Wu, Y., Guo, S., Zhang, L., Sun, H., 2020. Rose pattern for heliostat field
Energy 135, 479–486. https://doi.org/10.1016/j.solener.2016.05.060. optimization with a dynamic speciation-based mutation differential evolution. Int. J.
Bonanos, A.M., Blanco, M.J., Milidonis, K., 2019. Characterization of mirror soiling in Energy Res. 44 (3), 1951–1970. https://doi.org/10.1002/er.5048.
csp applications. AIP Conf. Proc. A. Dobbs, T. Elgindy, B.M. Hodge, and A. Florita. Short-term solar forecasting
T.E. Boukelia, A. Ghellab, A. Laouafi, and A. Cooling performances time series of csp performance of popular machine learning algorithms. Technical Report NREL/CP-
plants: Calculation and analysis using regression and ann models. Renewable 5D00-70030, National Renewable Energy Laboratory, 2017.
Energy, 2020. doi: 10.1016/j.renene.2020.05.012. ISSN 0960-1481. Du, S., Ren, Q., He, Y.L., 2017. Optical and radiative properties analysis and optimization
R. Buck. Heliostat field layout using non-restricted optimization. In SolarPACES 2012, study of the gradually-varied volumetric solar receiver. Appl. Energy 207, 27–35.
Marrakech, Morocco, 2012. https://doi.org/10.1016/j.apenergy.2017.05.165. Transformative Innovations for a
Cabello, J.M., Cejudo, J.M., Luque, M., Ruiz, F., Deb, K., Tewari, R., 2011. Optimization Sustainable Future - Part II. ISSN 0306-2619.
of the size of a solar thermal electricity plant by means of genetic algorithms. Du, S., He, Y.L., Yang, W.W., Liu, Z.B., 2018. Optimization method for the porous
Renewable Energy 36 (11), 3146–3153. https://doi.org/10.1016/j. volumetric solar receiver coupling genetic algorithm and heat transfer analysis. Int.
renene.2011.03.018. J. Heat Mass Transf. 122, 383–390. https://doi.org/10.1016/j.
Carballo, J.A., Bonilla, J., Berenguel, M., Fernandez-Reche, J., Garcia, G., 2019. New ijheatmasstransfer.2018.01.120. ISSN 0017-9310.
approach for solar tracking systems based on computer vision, low cost hardware Elsheikh, A.H., Elaziz, M.A., 2018. Review on applications of particle swarm
and deep learning. Renewable Energy 133, 1158–1166. https://doi.org/10.1016/j. optimization in solar energy systems. Int. J. Environ. Sci. Technol. 16 (2),
renene.2018.08.101. ISSN 0960-1481. 1159–1170. https://doi.org/10.1007/s13762-018-1970-x.
Carballo, Jose A., Bonilla, Javier, Berenguel, Manuel, Fernández-Reche, Jesús, Ammar H. Elsheikh, Swellam W. Sharshir, Mohamed Abd Elaziz, A.E. Kabeel, Wang
García, Ginés, 2019. Machine learning for solar trackers. AIP Conf. Proc. 2126 (1), Guilan, and Zhang Haiou. Modeling of solar energy systems using artificial neural
030012. https://doi.org/10.1063/1.5117524. network: A comprehensive review. Solar Energy, 180: 622–639, 2019. doi: 10.1016/
Carballo, Jose A., Bonilla, Javier, Berenguel, Manuel, Fernández, Jesús, García, Ginés, j.solener.2019.01.037. https://www.sciencedirect.com/science/article/pii/
2020. Solar tower power mockup for the assessment of advanced control techniques. S0038092X19300465. ISSN 0038-092X.
Renewable Energy 149, 682–690. https://doi.org/10.1016/j.renene.2019.12.075. A. Farahmand, S. Payan, and S.M. Hosseini Sarvari. Geometric optimization of radiative
ISSN 0960-1481. enclosures using PSO algorithm. International Journal of Thermal Sciences, 60:
Carrizosa, E., Dominguez-Bravo, C., Fernandez-Cara, E., Quero, M., 2015a. A heuristic 61–69, Oct 2012. doi:10.1016/j.ijthermalsci.2012.04.024.
method for simultaneous tower and pattern-free field optimization on solar power Farges, O., Bézian, J.J., El-Hafi, M., Fudym, O., Bru, H., 2013. Particle swarm
systems. Computers & OR 57, 109–122. optimization of solar central receiver systems from a monte carlo direct model. In:
Carrizosa, E., Domínguez-Bravo, C., Fernández-Cara, E., Quero, M., 2015b. Optimization Fudym, O., Battaglia, J.-L., Dulikravich, G.S., et al. (Eds.), IPDO 2013: 4th Inverse
of multiple receivers solar power tower systems. Energy 90, 2085–2093. https://doi. problems, design and optimization symposium. Albi, France.
org/10.1016/j.energy.2015.08.005. ISSN 0360-5442. Farges, O., Bezian, J.J., El-Hafi, M., 2018. Global optimization of solar power tower
Carrizosa, E., Dominguez-Bravo, C.A., Fernandez-Cara, E., Quero, M., 2017. An systems using a monte carlo algorithm: Application to a redesign of the ps10 solar
optimization tool to design the field of a solar power tower plant allowing heliostats thermal power plant. Renewable Energy 119, 345–353. https://doi.org/10.1016/j.
of different sizes. Int. J. Energy Res. 41 (8), 1096–1107. https://doi.org/10.1002/ renene.2017.12.028. ISSN 0960-1481.
er.3684. Fischer, S., Frey, P., Druck, H., 2012. A comparison between state-of-the-art and neural
Cavallaro, F., 2010. Fuzzy topsis approach for assessing thermal-energy storage in network modelling of solar collectors. Sol. Energy 86 (11), 3268–3277. https://doi.
concentrated solar power (csp) systems. Appl. Energy 87 (2), 496–503. https://doi. org/10.1016/j.solener.2012.09.002.
org/10.1016/j.apenergy.2009.07.009. ISSN 0306-2619. Freeman, J., Kiranlal, E.U., Dharmana, M.M., 2014. Novel anfis based control for solar
Chaturvedi, D.K., Isha, I., 2016. Solar power forecasting: A review. International Journal energy heliostats. Applied Mechanics and Materials 704, 395–400. https://doi.org/
of Computer Applications 145 (6), 28–50. https://doi.org/10.5120/ 10.4028/www.scientific.net/AMM.704.395.
ijca2016910728. Garcia-Martin, F.J., Berenguel, M., Valverde, A., Camacho, E.F., 1999. Heuristic
Cheng, Z., He, Y., Du, B., Wang, K., Liang, Q., 2015. Geometric optimization on optical knowledge-based heliostat field control for the optimization of the temperature
performance of parabolic trough solar collector systems using particle swarm distribution in a volumetric receiver. Sol. Energy 66 (5), 355–369. https://doi.org/
optimization algorithm. Appl. Energy 148, 282–293. https://doi.org/10.1016/j. 10.1016/S0038-092X(99)00024-9. ISSN 0038-092X.
apenergy.2015.03.079. ISSN 0306-2619. Ghritlahre, H.K., Prasad, R.K., 2018. Application of ANN technique to predict the
Cirocco, L., Belusko, M., Bruno, F., Boland, J., Pudney, P., 2014. Optimisation of storage performance of solar collector systems - a review. Renew. Sustain. Energy Rev. 84,
for concentrated solar power plants. Challenges 5 (2), 473–503. https://doi.org/ 75–88. https://doi.org/10.1016/j.rser.2018.01.001.
10.3390/challe5020473. ISSN 2078-1547. M. Gonzalez-Tokman, R. Avila-Miranda, and E. Sanchez. Neural control for a field of
Collado, F.J., Guallar, J., 2012. Campo: Generation of regular heliostat fields. Renewable concentrator heliostats. World Automation Congress Proceedings, pages 670–674,
Energy 46, 49–59. https://doi.org/10.1016/j.renene.2012.03.011. ISSN 0960-1481. 10 2014. doi:10.1109/WAC.2014.6936094.
Collado, F.J., Guallar, J., 2013. A review of optimized design layouts for solar power Guangyu, L., Zhongkun, C., 2017. Heliostat attitude angle detection method based on bp
tower plants with campo code. Renew. Sustain. Energy Rev. 20, 142–154. https:// neural network. MATEC Web of Conferences 139, 00043. https://doi.org/10.1051/
doi.org/10.1016/j.rser.2012.11.076. ISSN 1364-0321. matecconf/201713900043.
Collado, F.J., Guallar, J., 2018. Fast and reliable flux map on cylindrical receivers. Sol. Huang, Y., Wu, B., Chen, C., Chang, C., Kuo, T., 2009. Solar tracking fuzzy control system
Energy 169, 556–564. https://doi.org/10.1016/j.solener.2018.05.037. design using fpga. Lecture Notes in Engineering and Computer Science 2176, 07.
Conceicao, R., Silva, H.G., Collares-Pereira, M., 2018. Csp mirror soiling characterization IRENA, 2012. Renewable energy technologies: Cost analysis series. Irena working paper,
and modeling. Sol. Energy Mater. Sol. Cells 185, 233–239. https://doi.org/10.1016/ volume 1: Power sector, issue 2/5. International Renewable Energy Agency.
j.solmat.2018.05.035. IRENA. Renewable Power Generation Costs in 2019. International Renewable Energy
Cruz, N.C., Domingo Álvarez, J., Redondo, J.L., Berenguel, M., Ortigosa, P.M., Agency (IRENA), 2020. ISBN 9789292602529. https://books.google.com.cy/books?
Klempous, R., 2019. Control and optimal management of a heliostat field for solar id=MmoDEAAAQBAJ.
power tower systems. In: 2019 IEEE 23rd International Conference on Intelligent M. Izygon, K. McMurtrie, and N. Vu. Particle swarm optimization of the layout of a
Engineering Systems (INES), pp. 000271–000274. https://doi.org/10.1109/ heliostat field. AIP Conference Proceedings, 2033, 2018. doi:10.1063/1.5067054.
INES46365.2019.9109493. S.K. Jha, J. Bilalovic, A. Jha, N. Patel, and H. Zhang. Renewable energy: Present research
Cruz, N.C., Alvarez, J.D., Redondo, J.L., Berenguel, M., Ortigosa, P.M., 2018. A two- and future scope of artificial intelligence. Renewable and Sustainable Energy
layered solution for automatic heliostat aiming. Eng. Appl. Artif. Intell. 72, 253–266. Reviews, 77: 297–317, sep 2017. doi:10.1016/j.rser.2017.04.018.
https://doi.org/10.1016/j.engappai.2018.04.014. ISSN 0952-1976.

513
K. Milidonis et al. Solar Energy 224 (2021) 500–515

N. Jirasuwankul and C. Manop. A lab-scale heliostat positioning control using fuzzy logic of the year. Energy Convers. Manage. 93, 406–413. https://doi.org/10.1016/j.
based stepper motor drive with micro step and multi-frequency mode. In 2017 IEEE enconman.2015.01.021.
International Conference on Fuzzy Systems (FUZZ-IEEE), pages 1–6, July 2017. doi: Monterreal, R., Jimenez, A., 2003. Search for a real model of heliostat optics using
10.1109/FUZZ-IEEE.2017.8015479. genetic algorithms. ISES Solar World Congress 01.
S. Kalogirou and A. Sencan. Solar Collectors and Panels. Solar Collectors and Panels, I. Moukhtar, A.A. Elbaset, A.Z.E. Dein, Y. Qudaih, E. Blagin, D. Uglanov, and Y. Mitani. A
Theory and Applications. IntechOpen, 2010. ISBN 9533071427. developed concentrated solar power model using artificial neural network
Amir H. Keshavarzzadeh, Pouria Ahmadi, and Marc A. Rosen. Technoeconomic and technique. In 2017 Nineteenth International Middle East Power Systems Conference
environmental optimization of a solar tower integrated energy system for freshwater (MEPCON), pages 1346–1351, Dec 2017. doi:10.1109/MEPCON.2017.8301357.
production. Journal of Cleaner Production, page 121760, 2020. doi: 10.1016/j. Moukhtar, I., Elbaset, A.A., El-Dein, A.Z., Qudaih, Y., Blagin, E., Uglanov, D., Mitani, Y.,
jclepro.2020.121760. ISSN 0959-6526. 2018. Electric power regulation and modeling of a central tower receiver power
Khosravi, A., Malekan, M., Pabon, J.J.G., Zhao, X., Assad, M.E.H., 2020. Design plant based on artificial neural network technique. Journal of Renewable and
parameter modelling of solar power tower system using adaptive neuro-fuzzy Sustainable Energy 10, 043706. https://doi.org/10.1063/1.5029898.
inference system optimized with a combination of genetic algorithm and teaching Mubiru, J., Banda, E.J.K.B., 2008. Estimation of monthly average daily global solar
learning-based optimization algorithm. Journal of Cleaner Production 244, 118904. irradiation using artificial neural networks. Sol. Energy 82 (2), 181–187. https://doi.
https://doi.org/10.1016/j.jclepro.2019.118904. ISSN 0959-6526. org/10.1016/j.solener.2007.06.003.
Kiwan, S., Khammash, A.L., 2018. Investigations into the spiral distribution of the H. Nolte, T. Bello-Ochende, and J. Meyer. Second law analysis of a parabolic trough
heliostat field in solar central tower system. Sol. Energy 164, 25–37. https://doi.org/ receiver tube for small scale solar thermal application. In Proceedings of the Fifth
10.1016/j.solener.2018.02.042. ISSN 0038-092X. International Conference on Applied Energy (ICAE2013), Pretoria, South Africa, 07
Suhil Kiwan and Saif Al Hamad. On Analyzing the Optical Performance of Solar Central 2013.
Tower Systems on Hillsides Using Biomimetic Spiral Distribution. Journal of Solar Noone, C.J., Torrilhon, M., Mitsos, A., 2012. Heliostat field optimization: A new
Energy Engineering, 141 (1), 09 2018. doi:10.1115/1.4041101. ISSN 0199-6231. computationally efficient model and biomimetic layout. Sol. Energy 86 (2),
Suhil Kiwan and Abdel Latif Khammash. Optical Performance of a Novel Two-Receiver 792–803. https://doi.org/10.1016/j.solener.2011.12.007. ISSN 0038-092X.
Solar Central Tower System. Journal of Solar Energy Engineering, 142 (1), 07 2019. Diego Oliva, Mohamed Abd Elaziz, Ammar H. Elsheikh, and Ahmed A. Ewees. A review
doi:10.1115/1.4044189. ISSN 0199-6231. on meta-heuristics methods for estimating parameters of solar cells. Journal of
Kolb, G.J., Jones, S.A., Donnelly, M.W., Gorman, D., Thomas, R., Davenport, R., Power Sources, 435: 126683, 2019. doi: 10.1016/j.jpowsour.2019.05.089. https://
Lumia, R., 2007. Heliostat cost reduction study. Technical Report SAND2007-3293. www.sciencedirect.com/science/article/pii/S0378775319306548. ISSN 0378-7753.
Sandia National Laboratories, Albuquerque, New Mexico. Pai, N.S., Yau, H.T., Hung, T.H., Hung, C.P., 2013. Application of cmac neural network to
M.Y. Lazardjani, V. Kronhardt, G. Dikta, and J. Gottsche. Simultaneous optimization of solar energy heliostat field fault diagnosis. Int. J. Photoenergy 2013, 8. https://doi.
micro-heliostat geometry and field layout using a genetic algorithm. AIP Conference org/10.1155/2013/938162, 938162.
Proceedings, 1734, 2016. doi:10.1063/1.4949052. Pfahl, A., 2014. Survey of heliostat concepts for cost reduction. J. Sol.Energy Eng. 136
Lee, Kyungeun, Lee, Ikjin, 2019. Optimization of a heliostat field site in central receiver (1), 014501. https://doi.org/10.1115/1.4024243.
systems based on analysis of site slope effect. Sol. Energy 193, 175–183. https://doi. Picotti, Giovanni, Moretti, Luca, Cholette, Michael E., Binotti, Marco,
org/10.1016/j.solener.2019.09.027. Simonetti, Riccardo, Martelli, Emanuele, Steinberg, Theodore A.,
Levenberg, Kenneth, 1944. A method for the solution of certain non-linear problems in Manzolini, Giampaolo, 2020. Optimization of cleaning strategies for heliostat fields
least squares. Quart. Appl. Math. 2, 164–168. https://doi.org/10.1090/qam/10666. in solar tower plants. Sol. Energy 204, 501–514. https://doi.org/10.1016/j.
C. Li, R. Zhai, and Y. Yang. Optimization of a heliostat field layout on annual basis using solener.2020.04.032 http://www.sciencedirect.com/science/article/pii/
a hybrid algorithm combining particle swarm optimization algorithm and genetic S0038092X20304072. ISSN 0038-092X.
algorithm. Energies, 10 (11), 2017. doi:10.3390/en10111924. ISSN 1996-1073. Piroozmand, P., Boroushaki, M., 2016. A computational method for optimal design of the
Li, C., Zhai, R., Liu, H., Yang, Y., Wu, H., 2018. Optimization of a heliostat field layout multi-tower heliostat field considering heliostats interactions. Energy 106, 240–252.
using hybrid PSO-GA algorithm. Appl. Therm. Eng. 128, 33–41. https://doi.org/ https://doi.org/10.1016/j.energy.2016.03.049. ISSN 0360-5442.
10.1016/j.applthermaleng.2017.08.164. Pitz-Paal, R., Botero, N.B., Steinfeld, A., 2011. Heliostat field layout optimization for
Liggesmeyer, Peter, Rombach, Dieter, Bomarius, Frank, 2019. Smart Energy. Springer, high-temperature solar thermochemical processing. Sol. Energy 85 (2), 334–343.
Berlin Heidelberg, Berlin, Heidelberg, pp. 335–351. https://doi.org/10.1007/978-3- https://doi.org/10.1016/j.solener.2010.11.018. ISSN 0038-092X.
662-58134-6_20. Qi, H., Sun, S.C., He, Z.Z., Ruan, S.T., Ruan, L.M., Tan, H.P., 2016. Optimization
Lipps, F.W., Vant-Hull, L.L., 1978. A cellwise method for the optimization of large central Algorithms - Methods and Applications, chapter Inverse Geometry Design of
receiver systems. Sol. Energy 20 (6), 505–516. https://doi.org/10.1016/0038-092X Radiative Enclosures Using Particle Swarm Optimization Algorithms. IntechOpen.
(78)90067-1. ISSN 0038-092X. https://doi.org/10.5772/62351.
Liu, Hongtao, Zhai, Rongrong, Jiaxin, Fu., Wang, Yulong, Yang, Yongping, 2019. Ramos, A., Ramos, F., 2012. Strategies in tower solar power plant optimization. Sol.
Optimization study of thermal-storage pv-csp integrated system based on ga-pso Energy 86 (9), 2536–2548. https://doi.org/10.1016/j.solener.2012.05.024. ISSN
algorithm. Sol. Energy 184, 391–409. https://doi.org/10.1016/j. 0038-092X.
solener.2019.04.017. ISSN 0038-092X. P. Richter, E. Abraham, and G. Morin. Optimisation of concentrating solar thermal power
M. Lopez-Martinez, M. Vargas, and F.R. Rubio. Vision-based system for the safe operation plants with neural networks. In A. Dobnikar, U. Lotric, and B. Ster, editors, Adaptive
of a solar power tower plan. In F.J. Garijo, J.C. Riquelme, and M. Toro, editors, and Natural Computing Algorithms, pages 190–199, Berlin, Heidelberg, 2011.
Advances in Artificial Intelligence — IBERAMIA 2002, pages 943–952, Berlin, Springer, Berlin Heidelberg. ISBN 978-3-642-20282-7.
Heidelberg, 2002. Springer, Berlin Heidelberg. ISBN 978-3-540-36131-2. P. Richter, M. Frank, and E. Abraham. Multi-objective optimization of solar tower
Gabriel López, Christian A. Gueymard, Juan Luis Bosch, Igor Rapp-Arrarás, Joaquín heliostat fields. In Giovanni Russo, Vincenzo Capasso, Giuseppe Nicosia, and Vittorio
Alonso-Montesinos, Inmaculada Pulido-Calvo, Jesús Ballestrín, Jesús Polo, and Romano, editors, Progress in Industrial Mathematics at ECMI 2014, pages 771–778,
Javier Barbero. Modeling water vapor impacts on the solar irradiance reaching the Cham, 2016a. Springer International Publishing. ISBN 978-3-319-23413-7.
receiver of a solar tower plant by means of artificial neural networks. Solar Energy, P. Richter, D. Laukamp, L. Gerdes, M. Frank, and E. Abraham. Heliostat field layout
169: 34–39, 2018. doi: 10.1016/j.solener.2018.04.023. ISSN 0038-092X. optimization with evolutionary algorithms. In 2nd Global Conference on Artificial
Luo, L., Du, W., Wang, S., Wang, L., Sunden, B., Zhang, X., 2017. Multi-objective Intelligence (GCAI), 09 2016b. doi:10.29007/7p6t.
optimization of a solar receiver considering both the dimple/protrusion depth and Romero, M., Buck, R., Pacheco, J.E., 2002. An update on solar central receiver systems,
delta-winglet vortex generators. Energy 137, 1–19. https://doi.org/10.1016/j. projects, and technologies. J. Sol.Energy Eng. 124 (2), 98–108. https://doi.org/
energy.2017.07.001. ISSN 0360-5442. 10.1115/1.1467921.
Luo, Y., Lu, T., Du, X., 2018. Novel optimization design strategy for solar power tower Rubio, F.R., Berenguel, M., Camacho, E.F., 1995. Fuzzy logic control of a solar power
plants. Energy Convers. Manage. 177, 682–692. https://doi.org/10.1016/j. plant. IEEE Trans. Fuzzy Syst. 3 (4), 459–468. https://doi.org/10.1109/91.481955.
enconman.2018.09.089. ISSN 0196-8904. Salgado-Plasencia, Carrillo-Serrano, Rivas-Araiza, and Toledano-Ayala. Scada-based
Luo, Yan, Yue, Hu., Lu, T., 2019. Efficient optimized design of solar power tower plants heliostat control system with a fuzzy logic controller for the heliostat orientation.
based on successive response surface methodology. International Journal of Low- Applied Sciences, 9 (15): 2966, Jul 2019. doi:10.3390/app9152966. ISSN 2076-
carbon Technologies 14, 475–486. 3417.
Lutchman, S.L., Groenwold, A.A., Gauche, P., Bode, S., 2014. On using a gradient-based Salgado-Plasencia, Eugenio, Carrillo-Serrano, Roberto V., Toledano-Ayala, Manuel,
method for heliostat field layout optimization. Energy Procedia 49, 1429–1438. 2020. Development of a dsp microcontroller-based fuzzy logic controller for
https://doi.org/10.1016/j.egypro.2014.03.152. Proceedings of the SolarPACES heliostat orientation control. Applied Sciences 10 (5), 1598. https://doi.org/
2013 International Conference. ISSN 1876-6102. 10.3390/app10051598. ISSN 2076-3417.
Marquardt, Donald W., 1963. An algorithm for least-squares estimation of nonlinear Sanchez, M., Romero, M., 2006. Methodology for generation of heliostat field layout in
parameters. SIAM Journal on Applied Mathematics 11 (2), 431–441. https://doi. central receiver systems based on yearly normalized energy surfaces. Sol. Energy 80
org/10.1137/0111030. (7), 861–874. https://doi.org/10.1016/j.solener.2005.05.014. ISSN 0038-092X.
O. May, L.J. Ricalde, B. Ali, E. Ordoñez López, E. Venegas-Reyes, and O.A. Jaramillo. Sarvari, S.M.H., 2007. Optimal geometry design of radiative enclosures using the genetic
Neural network inverse modeling for optimization. In Artificial Neural Networks - algorithm. Numerical Heat Transfer, Part A: Applications 52 (2), 127–143. https://
Models and Applications. InTech, Oct 2016. doi:10.5772/63678. doi.org/10.1080/10407780601115020.
Mellit, A., 2008. Artificial intelligence technique for modelling and forecasting of solar Schlipf, D., Schneider, G., Maier, H., 2014. Using evolutionary algorithm to develop a
radiation data: a review. International Journal of Artificial Intelligence and Soft feed forward control for csp plant using mid- and long term storages. Energy
Computing 1 (1), 52–76. https://doi.org/10.1504/IJAISC.2008.021264. Procedia 49, 2191–2200. https://doi.org/10.1016/j.egypro.2014.03.232.
Mohammadi, K., Shamshirband, S., Tong, C.W., Alam, K.A., Petković, D., 2015. Potential Proceedings of the SolarPACES 2013 International Conference. ISSN 1876-6102.
of adaptive neuro-fuzzy system for prediction of daily global solar radiation by day

514
K. Milidonis et al. Solar Energy 224 (2021) 500–515

Schöttl, P., Rohani, S., Leonardi, E., Pisani, L., Les, I., Mutuberria, A., Nitz, P., 2019. Solar sorting genetic algorithm. Appl. Energy 205, 399–416. https://doi.org/10.1016/j.
field heliostat selection based on polygon optimization and boundaries. AIP Conf. apenergy.2017.07.096. ISSN 0306-2619.
Proc. 2126 (1), 030053. https://doi.org/10.1063/1.5117565. Wang, Pei, 2019. On defining artificial intelligence. Journal of Artificial General
Peter Schöttl, Gregor Bern, De Wet van Rooyen, José Antonio Fernández Pretel, Thomas Intelligence 10, 1–37.
Fluri, and Peter Nitz. Optimization of solar tower molten salt cavity receivers for Wang, X., Duan, Y., Xiao, B., Zhou, Z., Peng, H., Wang, Y., 2019. Hybrid solar power
maximum yield based on annual performance assessment. Solar Energy, 199: system optimization based on multi-objective pso algorithm. In: 2019 Chinese
278–294, 2020. doi: 10.1016/j.solener.2020.02.007. ISSN 0038-092X. Automation Congress (CAC), pp. 4176–4180. https://doi.org/10.1109/
T. Senjyu, S. Chakraborty, A.Y. Saber, H. Toyama, A. Yona, and T. Funabashi. Thermal CAC48633.2019.8997401.
unit commitment strategy with solar and wind energy systems using genetic Wei, X., Lu, Z., Wang, Z., Yu, W., Zhang, H., Yao, Z., 2010. A new method for the design
algorithm operated particle swarm optimization. In 2008 IEEE 2nd International of the heliostat field layout for solar tower power plant. Renewable Energy 35 (9),
Power and Energy Conference. IEEE, Dec 2008. doi:10.1109/pecon.2008.4762597. 1970–1975. https://doi.org/10.1016/j.renene.2010.01.026. ISSN 0960-1481.
Jyoti Sharma and Ravi Shankar Singhal. Comparative research on genetic algorithm, Xu, J., Liu, Y., Luo, Y., Ye, W., Wang, Z., 2017. Control strategy of outlet steam
particle swarm optimization and hybrid ga-pso. In 2015 2nd International temperature for dsg in parabolic troughs solar power under recirculation operation
Conference on Computing for Sustainable Global Development (INDIACom), pages mode based on fuzzy dynamic matrix control. In: 2017 2nd International Conference
110–114, 2015. on New Energy and Renewable Resources (ICNERR 2017), 2nd International
Y. Song, W. Huang, and X. Zhu. A vision-based fault diagnosis system for heliostats in a Conference on New Energy and Renewable Resources. ISBN 978-1-60595-470-7.
central receiver solar power plant. In Proceedings of the 10th World Congress on Yadav, A.K., Chandel, S.S., 2014. Solar radiation prediction using artificial neural
Intelligent Control and Automation, pages 3417–3421, July 2012. doi:10.1109/ network techniques: A review. Renew. Sustain. Energy Rev. 33, 772–781. https://
WCICA.2012.6359038. doi.org/10.1016/j.rser.2013.08.055.
Soylemez, M.S., Aldulaimi, R.K.M., 2016. An innovative optimization algorithm for Yaici, W., Entchev, E., 2016. Adaptive neuro-fuzzy inference system modelling for
précis estimation of the optical efficiency of heliostat fields. International Journal of performance prediction of solar thermal energy system. Renewable Energy 86,
Renewable Energy Research 6 (3). 302–315. https://doi.org/10.1016/j.renene.2015.08.028.
S. Sun, H. Qi, Z. Fang-Zhou, R. Li-Ming, and L. Bing-Xi. Inverse geometry design of two- Yao, Y., Hu, Y., Gao, S., 2015. Heliostat field layout methodology in central receiver
dimensional complex radiative enclosures using krill herd optimization algorithm. systems based on efficiency-related distribution. Sol. Energy 117, 114–124. https://
Applied Thermal Engineering, 98: 1104–1115, 01 2016. doi:10.1016/j. doi.org/10.1016/j.solener.2015.04.029. ISSN 0038-092X.
applthermaleng.2016.01.017. Youssef, A., El-Telbany, M., Zekry, A., 2017. The role of artificial intelligence in photo-
P. Talebizadeh, M.A. Mehrabian, and H. Rahimzadeh. Optimization of heliostat layout in voltaic systems design and control: A review. Renew. Sustain. Energy Rev. 78,
central receiver solar power plants. Journal of Energy Engineering, 140 (4), 2014. 72–79. https://doi.org/10.1016/j.rser.2017.04.046. ISSN 1364-0321.
doi:10.1061/(ASCE)EY.1943-7897.0000162. A. Zeghoudi and A. Chermitti. Estimation of the solar power tower heliostat position
Terhag, Felix, Wolfertstetter, Fabian, Wilbert, Stefan, Hirsch, Tobias, Schaudt, Oliver, using neural network. International Journal of Computer Applications, 94: 41–46, 05
2019. Optimization of cleaning strategies based on ann algorithms assessing the 2014. doi:10.5120/16335-5620.
benefit of soiling rate forecasts. AIP Conf. Proc. 2126 (1), 220005. https://doi.org/ A. Zeghoudi and A. Chermitti. Speed control of a dc motor for the orientation of a
10.1063/1.5117764 https://aip.scitation.org/doi/abs/10.1063/1.5117764. heliostat in a solar tower power plant using artificial intelligence systems (flc and
Wang, J., Duan, L., Yang, Y., 2018. An improvement crossover operation method in nc). Research Journal of Applied Sciences, Engineering and Technology, 10:
genetic algorithm and spatial optimization of heliostat field. Energy 155, 15–28. 570–580, 06 2015. doi:10.19026/rjaset.10.2465.
https://doi.org/10.1016/j.energy.2018.05.004. Zeghoudi, A., Chermitti, A., Benyoucef, B., 2016. Contribution to the control of the
Wang, Jianxing, Guo, Lili, Zhang, Chengying, Song, Lei, Duan, Jiangyong, Duan, Liqiang, heliostat motor of a solar tower power plant using intelligence controller. Int. J.
2020. Thermal power forecasting of solar power tower system by combining Fuzzy Syst. 18 (5), 741–750. https://doi.org/10.1007/s40815-015-0098-0.
mechanism modeling and deep learning method. Energy 208, 118403. https://doi. Zhang, M., Yang, L., Xu, C., Du, X., 2016. An efficient code to optimize the heliostat field
org/10.1016/j.energy.2020.118403 http://www.sciencedirect.com/science/article/ and comparisons between the biomimetic spiral and staggered layout. Renewable
pii/S0360544220315103. ISSN 0360-5442. Energy 87, 720–730. https://doi.org/10.1016/j.renene.2015.11.015. ISSN 0960-
Wang, K., He, Y.L., Li, P., Li, M.J., Tao, W.Q., 2017. Multi-objective optimization of the 1481.
solar absorptivity distribution inside a cavity solar receiver for solar power towers. Ruidi Zhu and Dong Ni. An optimized aiming strategy tracking flux set point for solar
Sol. Energy 158, 247–258. https://doi.org/10.1016/j.solener.2017.09.044. power tower system. 2019 Chinese Automation Congress (CAC), pages 1386–1391,
Wang, K., He, Y.L., Xue, X.D., Du, B.C., 2017. Multi-objective optimization of the aiming 2019.
strategy for the solar power tower with a cavity receiver by using the non-dominated

515