You are on page 1of 17

Solar Energy 202 (2020) 210–226

Contents lists available at ScienceDirect

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

Techno-economic optimization of grid-connected, ground-mounted T


photovoltaic power plants by genetic algorithm based on a comprehensive
mathematical model
Martin János Mayer , Gyula Gróf

Department of Energy Engineering, Faculty of Mechanical Engineering, Budapest University of Technology, Műegyetem rkp. 3, Budapest 1111, Hungary

ARTICLE INFO ABSTRACT

Keywords: The increasing penetration of photovoltaic (PV) technology calls for the development of an effective method for
Grid-connected photovoltaic plants optimization of grid-connected photovoltaic power plants. This paper presents a simultaneous optimization
Modeling method of ten important design parameters of a PV plant, including the module power, inverter sizing, support
Optimization structure dimensions, cable losses, module orientation and row spacing. A mathematical PV performance model
Genetic algorithm
taking into account the important effects and losses and an economic cost model were developed and presented
in detail. The objective function is the internal rate of return and the optimization is performed by a genetic
algorithm. The results show that the proposed models and method are capable to optimize the grid-connected PV
plant and provide reliable results after a 6–7 min calculation time. The method was demonstrated in detail for a
Hungarian location, including the losses and cost structure of the optimal plant configuration. The optimization
was also performed for 5 additional sites around the world to assess the effect of location and meteorology. The
impact of the decreasing PV module prices on the optimal design is calculated to identify the expected future
trends in PV plant design. The presented optimization method can be utilized to facilitate the optimal design of
commercial PV plants and for research purposes.

1. Introduction topic deal with standalone PV systems. The design of standalone sys-
tems is largely affected by several aspects, like the load profile, storage
Photovoltaic (PV) systems – producing electricity directly from the and possibilities of hybridization, which are not important for large-
solar radiation – are one of the most promising ways to ensure the clean scale GCPV plants. These papers are not discussed in detail as even the
and sustainable power supply of mankind. As PV is the fastest-growing formulation of the optimization problem is different for standalone and
renewable energy technology with a 32% average growth rate between grid-connected systems. The relatively small number papers dealing
2012 and 2017 in terms of installed capacity, developing easy-to-use with GCPVs typically aim to find the optimal type and number of PV
methodologies to enhance the optimal design of these power plants has modules, the inverter sizing ratio and the tilt angle.
a high importance. Driven by the decreasing cost of the technology, the Kornelakis and Marinakis (2010) used particle swarm optimization
different subsidies and the availability of low-cost capital, large-scale (PSO) for optimal sizing of GCPV plants. The decision variables are PV
grid-connected photovoltaic (GCPV) power plants became the most modules’ optimal number, tilt angle, placement on the mounting
prevalent PV project types with 77% share in worldwide installed ca- structures and distribution among the inverters. The objective function
pacity in 2017 (Renewable Energy Policy Network for the 21st Century, is the net present value (NPV) over the lifetime of the plant, taking into
2018). These power plants are mainly designed using intuitive and accounts the estimated installation and maintenance costs and revenues
analytical methods, which often fail to find the optimal plant config- from the produced energy. The model contains a detailed description of
uration. This paper focus on the design optimization of ground- the mounting structure, but neglects several losses like shading and
mounted GCPV power plants based on extensive numeric modelling, cable losses, which would also affect the optimal layout. Kornelakis
including both technical and economic aspects. (2010) expands the previously described optimization by adding the
Khatib et al. (2013) composed a review on the optimization of the avoided CO2 emission as a second objective function. The calculations
PV systems size, which highlighted that most research papers in the rely on the same model and methodology as in the previous paper. The


Corresponding author.
E-mail address: mayer@energia.bme.hu (M.J. Mayer).

https://doi.org/10.1016/j.solener.2020.03.109
Received 8 April 2019; Received in revised form 5 March 2020; Accepted 28 March 2020
0038-092X/ © 2020 International Solar Energy Society. Published by Elsevier Ltd. All rights reserved.
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

Nomenclature H0 bottom of the pile foundation from the bottom of the


modules, m
α inter-row spacing angle, ° hane height of wind speed data in the meteorological database
αc absorption coefficient of the solar cell hmod average distance of modules from the ground
αO&M operation and maintenance cost factor I current, A
αS solar elevation angle, ° I0 diode reverse saturation current, A
β tilt/slope/inclination angle of the PV modules, ° IL light current, A
γ azimuth angle of the PV modules (east: −90°, south: 0°, Imp MPP current of the PV module, A
west: 90°), ° Isc short circuit current of the PV module, A
γS solar azimuth angle, ° IRR internal rate of return
ε band gap of the semiconductor, 1.121 eV for silicon cells k Boltzmann’s constant, 1.381∙10−23 JK−1
ηinv inverter efficiency Kθ incidence angle modifier
ηrel,200 relative efficiency of PV module at 200 W/m2 ltr loss factor of transformers and switchgears
ηSTC PV module efficiency at STC ldeg yearly degradation rate
θZ solar zenith angle, ° lLID light-induced degradation loss factor
θ incidence angle of the solar radiation, ° lsoil soiling loss factor
μI,sc temperature coefficient of the PV module’s short circuit Lmod length of a PV module, m
current at STC, K−1 Lseg length of one segment of support structure lines, m
μP,mp temperature coefficient of the PV module’s maximum Lst total length of the support structure lines, m
power at STC, K−1 MTFBinv mean time between failures of the inverters, a
μV,oc temperature coefficient of the PV module’s open circuit n diode ideality factor
voltage at STC, K−1 Ncg number of cell groups separated by bypass diodes along a
ρCu resistivity of copper, 1.68∙10−8 Ωm structure line width
ρg ground albedo Ninv number of inverters
τ transmission coefficient of the glass cover Nmod number of PV modules
a modified diode ideality factor, V Nmpl number of module rows in one mounting structure line
aland land area extension factor Ns number of modules connected series in a string
Aland land area of the GCPV plant, m2 Nstr total number of strings
Al,0 land area of the control building, m2 Nsc number of series cells in the PV module
Al,mod land area covered by the modules, m2 Np number of strings connected parallel to an inverter
b0 incidence angle modifier coefficient NPV net present value, €
Bh beam (direct) horizontal irradiance, Wm−2 pel electricity unit rate, €kWh−1
Bt beam (direct) irradiance on a tilted surface, Wm−2 PAC AC output power of an inverter, W
Btot total equivalent length of the support structure spars, m PDC DC input power of an inverter, W
Bvert length of the vertical spars in one segment of the support Pgrid total power fed into the grid, W
structure, m Pmp MPP power of a PV module, W
Cc,AC total installation cost of low voltage AC cables, € r discount rate
Cc,DC total installation cost of DC string cables, € rb beam radiation transposition factor
Cinst total installation cost of the GCPV plant, € rd diffuse transposition factor
Cinv total installation cost of the inverters, € rd,LJ view factor of the isotropic diffuse radiation
Cland total cost of land area and lightning protection, € rd,LJ,mod modified view factor of the isotropic diffuse radiation
CO&M annual operation and maintenance cost, €/a considering row shading
CPV total installation cost of the PV modules, € rr reflected radiation transposition factor
Crepl annual inverter replacement cost, €/a rtax taxation rate
Cstruct total installation cost support structures, € R annual revenues of the power plant, €/a
Ctr total installation cost of transformers and switchgears, € Rc,DC average resistance of DC cabling of one string, Ω
cc,AC unit cost of low voltage AC cables, €m−1 Rc,AC average resistance of AC cabling of one inverter, Ω
cc,DC unit cost of DC cables, €m−1 Rs series resistance of the PV module, Ω
cinv unit cost of one inverter, € Rsh shunt resistance of the PV module, Ω
cland unit cost of land area, €m−2 Rt reflected on a tilted surface, Wm−2
cmod unit cost of one PV module, € S upfront incentives, €
cstruct unit cost of support structure spars, €m−1 Δt timestep of meteorological database, h
Dh diffuse horizontal irradiance, Wm−2 Ta ambient temperature, °C
Dt diffuse irradiance on a tilted surface, Wm−2 Tc cell temperature, °C
D1 horizontal width a support structure row, m Tc,STC cell temperature at STC, 25 °C
D2 inter-row spacing, m tlt design lifetime of the plant, a
Dmin minimum inter-row spacing, m V voltage, V
Gh global horizontal irradiance, Wm−2 VDC DC input voltage of an inverter, V
GPV global irradiance on the unshaded surface of PV modules, Vl line voltage of low voltage AC network, 400 V
Wm−2 Voc open circuit voltage of a PV module, V
GPV,sh global irradiance on the shaded surface of PV modules, ΔvAC relative nominal voltage drop on AC cables
Wm−2 ΔvDC relative nominal voltage drop on DC cables
Gt global irradiance on a tilted surface, Wm−2 v wind speed, ms−1
GSTC radiation at STC, 1000 Wm−2 vane wind speed data in the meteorological database, ms−1
H height of the modules on the support structure row vmod wind speed at the height of the modules, ms−1

211
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

W width of the modules on the support structure row support structure, m


Wmod width of a PV module, m wsh shaded proportion of a support structure row
Wstand maximum width supported by one vertical spar of the

results show the tradeoff between economic and environmental aspects Sulaiman et al. (2012) dealt with the choice of module and inverter
during the GCPV plant design. and the optimization of the number of modules connected in series and
Gómez et al. (2010) optimized the PV system size and location in parallel in a rooftop GCPV system. Firstly, an iterative sizing algorithm
order to maximize the profitability index of the plant by binary PSO and is introduced based on analytical calculations, which proved to be ef-
genetic algorithm (GA). The applied PV system model is very simple as fective, but slower than the later proposed evolution algorithms. The
it calculates only with a constant, location-dependent loss factor, the optimization is performed for both a technical and an economic ob-
paper focused on the comparison of the different optimization methods. jective function, the specific yield and the NPV respectively, resulting in
The results show the slight supremacy of the proposed binary PSO al- largely different results, which highlighted the importance of the choice
gorithm, but the performance of GA is also satisfying. PSO was also of a proper objective.
proved to be suitable in optimizing conventional power plants in Notton et al. (2010) presented an optimization of the inverter sizing
(Groniewsky, 2013). ratio by maximizing the total efficiency of the modules and the inverter.
Gómez-Lorente et al. (2012) compared four evolutionary algorithms The effect of the PV module technology (m-Si, p-Si, a-Si, CIS), the in-
(EA) in optimizing solar tracking PV plants. The only objective function clination angle, the location and the inverter efficiency curve on the
is the electric loss of the cables, the energy production and the other optimal sizing ratio is analyzed. The results show that the efficiency
losses of the modules and the costs of the components are not con- curve has the most influence on the optimum, but all other parameters
sidered. Differential evolution (DE) proves to be the best among the affect it to some extent.
assessed algorithms for this purpose. The calculation is extended to a Chen et al. (2013) presented a more detailed optimization of the
multi-objective optimization problem in (Gómez-Lorente et al., 2014) inverter sizing with an objective function of the total cost savings
by adding the maximization of installed capacity as a second objective. compared to base case, where the total nominal power of the inverters
A more detailed analysis of tracking systems is presented in Bakhshi and PV modules are equal. The calculations aim to determine the effect
and Sadeh (2016), which compared the fixed, horizontal and vertical of the meteorological conditions of the location, the economic in-
single axis and dual axis tracking systems. The objective is to maximize centives and electricity rates and the inverter parameters and overload
the profitability expressed by the NPV, internal rate of return (IRR) and protection on the optimal inverter sizing. The comparison of sites in Las
payback time (PBT) parameters. The annual energy production is cal- Vegas and Eugene reveals that the typically higher irradiance in Las
culated from hourly meteorological data using a model describing the Vegas is counteracted by the higher temperature in terms of the fre-
radiation on the tilted surface, the module temperature and perfor- quency of high PV power output, which makes the optimal inverter size
mance and the losses of the inverters. However, the cost of land area similar. The increase of electricity rates and incentives proportional to
and shading losses are neglected, and all the other losses are summar- the DC power encourage the undersizing of the inverter, while in-
ized in a constant cable loss parameter. The trackers with single vertical centives proportional to AC power have an opposite effect. The minimal
axis are most profitable for the examined location and conditions. startup power, the part-load efficiency and the overload protection
Bakhshi et al. (2014) introduced a method to optimize a large-scale scheme of the inverter are all influence the optimal sizing. The results
rooftop PV system layout for a given location using genetic algorithm. show that the use of a detailed and accurate model for the optimization
The decision variables are the number of inverters, the number of used is crucial, since neglecting some effect can largely influence the optimal
inputs per inverter, number of parallel strings per input and the number properties. Nonetheless, the incidence angle of the modules, the
of series modules per string. The objective function is the NPV, calcu- shading, cable and other electrical losses are not included in the model.
lated by the costs and revenues of the system. The model used for the Martins Deschamps and Rüther (2019) optimized the inverter sizing
calculations is similar to the one presented in Bakhshi and Sadeh based on real measured production data of 5 PV technologies (a-Si/μc-
(2016). The system optimized with the proposed method has slightly Si, a-Si, CIGS, p-Si, m-Si). The optimal sizing largely depends on the DC
better economic parameters as the one that actually has been installed. cost ratio, i.e. the share of the cost of all DC-related components among
Bakhshi-Jafarabadi et al. (2019) used binary linear programming for the whole installation cost, and there are also major differences be-
the same optimization problem, which has proved to be faster and more tween the PV modules types. The calculation based on measured data
accurate than evolutionary algorithms in finding the global optimum. improves the reliability of the results but makes it impossible to conduct

Fig. 1. Design optimization process of grid-connected photovoltaic power plants.

212
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

a proper optimization since the effect of different factors can only be sensitivity analysis, while is also presents an assessment of the effect of
considered by modelling. location and meteorology, the impact of reducing PV prices and the
Aronescu and Appelbaum (2017) optimized the solar field of GCPV boundary conditions and further application possibilities of the method.
plants with the decision variables of the inclination angle, the distance Section 7 highlights the main conclusions and the further research
between the rows, the height and number of the rows, the number of concepts.
modules connected in series and parallel and the length and width of
the total field area. The objective functions are the maximum annual 2. Optimization problem
incident energy, minimum field area, minimum plant cost and
minimum cost of unit energy, of which only the last one can cover all The proposed process of the GCPV optimization algorithm is sum-
the trade-offs and provide realistic results. marized in Fig. 1. The ten decision variables are among the most im-
Vokony et al. (2018) focused on the optimization of the tilt angle of portant design parameters of GCPV power plants. The first variable is
PV modules, highlighting the importance of considering all relevant the nominal power of a PV module. Modules with higher nominal
factors, e.g. shading, dust, wind load and construction cost during op- power produce more energy but they are also more expensive even
timization. It claims the real optimal tilt angle in Europe to be between specified for a unit of power. The number of series modules in a string
25 and 28°, lower than the 36° which gives the yearly maximum in- and the number of parallel strings connecting to an inverter, together
plane radiation. with the module power, determine the AC to DC power ratio of the
Perez-Gallardo et al. (2014) presented an optimization of GCPV plant, the most important optimization parameter in the literature
plants by considering both economic and environmental aspects using (Khatib et al., 2013). Moreover, they also affect the cable and inverter
the objective functions of payback time (PBT) and energy payback time losses and the cable costs. The total number of inverters describe the
(EPBT) respectively. The environmental impacts are calculated using total AC power of the plant, which is also a trade-off between the re-
Life Cycle Assessment (LCA), while the optimization was performed latively large fixed cost components of a smaller plant and the extra
with a genetic algorithm. Among five PV technologies (m-Si, p-Si, a-Si, costs of finding a proper location and grid connection point for a larger
CdTe, CIS) the m-Si and the CdTe proves to be the best in terms of BPT plant. The width of the support structure influences the height and
and EPBT, respectively. The results highlight the difference between the material demand of the structure while also affects the shading losses.
economic and environmentally optimal design. The nominal voltage drops on DC and AC cables represent a trade-off
Despite the practical and scientific importance of the issue, only few between the cables costs and the power dissipation. The tilt angle and
papers deal with the design optimization of large-scale GCPV plants. azimuth determine the in-plane irradiance and, together with the inter-
Most of these studies only focus on one or several of the different as- row spacing, they also influence the shading losses, the total land area
pects of design, e.g. inverter sizing, module layout, tilt angle, row of the plant and the cable lengths. All design parameters represent
spacing, cable losses, but no method has yet developed that takes all compromises between the energy production and the costs of the power
important aspects into consideration. The global optimum of a large- plant, therefore both technical and economic performance has to be
scale GCPV plant can only be found by simultaneously optimizing all modelled and evaluated to find their optimal value. The IRR was chosen
the important parameters due to the interrelations between them. as the objective function as it is universal and not sensitive to uncertain
This paper introduces a comprehensive optimization of ten im- parameters, like the levelized cost of electricity (LCOE) on the discount
portant design parameters of a ground-mounted GCPV plant using ge- rate.
netic algorithm. The ten optimized design parameters are listed in The optimization results are only reliable when all significant effects
Fig. 1, while the objective function is the IRR, incorporating both the of the decision variables on both the technical performance and costs
technical and economic impact of the decision variables. Section 2 in- are considered in the models. In the other hand, the large computa-
troduces the optimization problem and the decision variables. The tional requirement of an overly complex model can hinder the appli-
mathematical model used to calculate the annual energy production of cation of a global optimization method as they typically involve a large
the plant as a function of the design parameters is described in detail in number of model evaluations. The established performance and cost
Section 3. The calculation of the installation and maintenance costs and models represent a compromise between accuracy and complexity and
the financial return of the system is presented in Section 4. Section 5 are able to be used in this extensive optimization problem. The models
describes the optimization process. Section 6 introduces the results for a are thoroughly described in Sections 3 and 4 to clearly present the used
sample system in Hungary with detailed losses and cost distribution, a methods and simplifications in the models.

Fig. 2. Flow of the PV plant technical performance calculation process, including all inputs (green), models and variables (blue), output (bourdon) and the direct
effect of the design parameters (red). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

213
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

3. Photovoltaic performance model calculates rd simply as a view factor to the sky:


1 + cos
The PV plant considered in this paper is a large-scale grid-connected rd, LJ =
2 (6)
photovoltaic (GCPV) power plant. The modules are mounted on mul-
tiple lines of support structures with the same size and orientation and The anisotropic models introduce a second, circumsolar component
uniform row spacing. The electric connection of the modules is homo- since, especially during sunny weather, a significant part of the diffuse
geneous, i.e. the same number of strings are connected to each inverter radiation is coming from the proximity of the sun disc. The most so-
and each string is composed by the same number of modules. The phisticated models also include the horizon brightening as a third
profitability of a GCPV plant basically depends on its annual energy component of the diffuse radiation. Gulin et al. (2013) compared the
production, which is typically calculated by computer simulation based results of 3 isotropic and 6 anisotropic models with real measured data
on meteorological databases. A mathematical model is used to calculate and the two-component Hay model proved to be the most accurate one.
the output power of the PV plant, considering all important factors The Hay model also performed among the bests in the comparison of 15
affecting the performance and losses of the system, as a function of the different models in Gracia and Huld (2013). The Hay model an FHay
meteorological variables. The input variables of the model are the anisotropy index is defined based on the beam irradiance and the G0
horizontal beam and diffuse radiation, the ambient temperature and the solar constant, which determine the circumsolar fraction of the diffuse
wind speed, while the output is the electric power fed into the grid, irradiance (Hay, 1993):
which is integrated over a year to get the annual energy yield. Covering Bh
all the effects of the decision variables in the model with an acceptable FHay =
G0cos Z (7)
accuracy is important for the optimization, since neglecting any of them
can lead to suboptimal solution and misleading conclusions. rd = FHay rb + (1 FHay ) rd, LJ (8)
Accordingly, the used model includes the calculation of the global ra-
diation on a tilted plane including the reflection, shading and soiling The circumsolar part is treated in the same way as the global irra-
losses, the cell temperature, the PV module power, the DC and AC cable diance, while the rest is considered isotropic and calculated according
losses and the inverter electric and overload losses. A flowchart of the to the Liu-Jordan model.
performance calculation process is shown in Fig. 2, while the detailed The reflected irradiance is much smaller than the other components
calculation is presented in the following sections. in practical cases, therefore it is typically calculated using a simple
equation. This assumes the same reflectivity, the ρg albedo of the
ground for both the direct and diffuse irradiance and consider a purely
3.1. Radiation on a tilted plane
isotropic reflection using the rr view factor from ground to the module
(Gracia and Huld, 2013):
Meteorological databases contain radiation data for the horizontal
plane, which must be transposed to the plane of the PV modules. The 1 cos
rr =
orientation of the modules is defined by the β tilt angle and the γ azi- 2 (9)
muth. The position of the Sun is described by the θZ zenith angle (its
Rt = g Gh rr (10)
complementary angle, the αS elevation angle can also be used) and the
γS solar azimuth angle. These angles are shown in Fig. 3. The solar The reflection of the radiation from the module surface is con-
angles are calculated based on the time and the geographical location sidered using a Kθ incidence angle modifier (IAM) calculated with the
using general astronomical equations described in Duffie and Beckman following equation (Duffie and Beckman, 2013):
(2013). The beam and diffuse components of the global horizontal ir-
radiance, in case they are not included in the database, can be calcu- K =1 b0
1
1
lated from the global irradiance with – among others – the empirical cos (11)
model introduced in Erbs et al. (1982).
The b0 incidence angle modifier coefficient has a typical value of
The Gt global irradiance on a tilted plane is the sum of the Bt beam,
0.05 for PV modules Sjerps-Koomen et al. (1996). The incidence angle
Dt diffuse and Rt reflected radiation components:
modifier is separately determined for the beam, diffuse and reflected
Gt = Bt + Dt + Rt (1) irradiance. The effective incidence angle of the diffuse and reflected
radiation is calculated with the following equation proposed by
The beam irradiance on a tilted surface, as it comes directly from
the Sun, is given by a purely geometrical equation. The rb beam
transposition factor is defined as (Gulin et al., 2013):
cos
rb =
cos Z (2)

Bt = Bh rb (3)

where θ, the incidence angle of the radiation on the tilted plane, cal-
culated from the angles of the Sun and the surface using the following
equation (Duffie and Beckman, 2013):
cos = cos Z cos + sin Z sin cos( S ) (4)

The diffuse irradiance on a tilted plane cannot be exactly calculated


based on the available data, therefore many different models exist to
estimate its value. The tilted diffuse irradiance is expressed using an rd
diffuse transposition factor, calculated differently in each model.
Dt = Dh rd (5)

The isotropic models assume that the diffuse radiation is uniform Fig. 3. Zenith (θZ), elevation (αS) and azimuth (γS) angles of the Sun and tilt (β)
throughout the sky dome. The purely isotropic Liu-Jordan model and azimuth (γ) angles of the PV modules.

214
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

Brandemuehl and Beckman (1980): GPV = {Bh rb K ,b + Dh [FHay rb K , + (1 FHay ) rd, LJ , mod K diff ] + g Gh rr K relf }

diff = 59.68 0.1388 + 0.001497 2


(12) (1 lsoil ) (17)

(13)
= 90 0.5788 + 0.002693 2 GPV , sh = [Dh (1 FHay ) rd, LJ , mod K +
refl diff g Gh rr K relf ](1 lsoil ) (18)
The effective diffuse incidence angle applies for isotropic diffuse The irradiance of the partially shaded cell at the border of the
radiation, therefore using it with the anisotropic Hay model can lead to shading is calculated as the average of GPV and GPV,sh weighted by the
inaccuracy (Strobach et al., 2013). To solve this, the first term of (8), unshaded and shaded area of the given cell, respectively.
referring to the circumsolar radiation is multiplied with the beam IAM The electric losses due to the hard shading depends on the or-
and only the second, isotropic diffuse term is corrected with the IAM ientation of the module and the number of modules along the W width
calculated using the effective diffuse incidence angle. of a support structure line. The proportion of electric shading losses is
The irradiance loss due to the dust accumulation of the modules generally higher than the decrease of the radiation since only one
depends on the tilt angle, therefore it must be considered in the model. shaded cell can limit the current of all the series connected cell and
Abdeen et al. (2017) and You et al. (2018) developed models to de- modules. For simplicity, the power of all cells in a string is considered
scribe dust accumulation over time, which can be even used for as the same as the lowest cell power. This power decrease can be re-
cleaning optimization. However, due to their complexity and parameter duced using bypass diodes, which can enable a higher current for the
requirements, the use of these models during design is only reasonable unshaded cell at the cost of the exclusion of the shaded cells from the
in deserted areas where soiling is an important issue. The correlation energy production. A common 60-cell crystalline Si PV module typi-
used in this work is derived based on the experimental results of Cano cally has 3 bypass diodes, dividing the module into three groups of 20
et al. (2014)), which is more effective for the purpose of the presented cells along its short edge as shown in Fig. 6. The Ncg number of in-
optimization: dependent cell groups along the width of a structure line equals to the
Nmpl number of module rows on a line in case of portrait module or-
lsoil = 0.01385 0.000155 (14)
ientation, and the Nmpl times three for landscape orientation.
This average soiling loss factor is considered constant over time and The PV characteristic has three local maxima in the presented case
applies for the global tilted irradiance. due to the three irradiance levels (Gallardo-Saavedra and Karlsson,
2018):
3.2. Shading losses
1) no bypass diodes activated, all cells produce energy with the power
The irradiance calculated according to the presented equations is limited by the shaded cells
valid to an individual PV module, but for multiple row PV arrangements 2) bypass diodes activated for shaded and partially shaded cells, only
shading is also an important effect to consider. The dimensions of the the unshaded cells produce electricity with their possible maximum
supporting structure lines are visible in Fig. 4. power
An important, but often overlooked effect is the decrease of the 3) bypass diodes activated for totally shaded cell, the unshaded and
diffuse radiation due to the soft shading of the adjacent lines, visualized partially shaded cells produce energy with the power enabled by the
by the arrows in Fig. 4. This effect can be calculated by the modification partially shaded cells
of the view factor, as stated in (Aronescu and Appelbaum, 2017). The
modified isotropic diffuse transposition factor for the lowest point of The activated bypass diodes have a ΔVbpd voltage drop, which is also
the modules is: considered in the latter two cases. A perfect maximum power point
tracking (MPPT) is assumed, which can always adjust the string voltage
1 + cos( + ) to find a global maximum of the three presented cases (Winston et al.,
rd, LJ , mod =
2 (15) 2018). The same shading conditions are assumed for all strings; thus no
This effect decreases from the bottom to the top of the modules. other losses are considered due to the voltage mismatch of strings. The
However, due to the series connection of modules the lowest irradiance lsh electric shading loss factor, derived by the presented method, is
limits the current of the whole string (Gallardo-Saavedra and Karlsson, shown as a function of the wsh relative shaded area in Fig. 7 for
2018). Therefore, calculating with (15) for the whole surface is a fair GPV = 800 W/m2, GPV,sh = 200 W/m2 and Nmpl = 2. Fig. 7a) applies for
approximation. modules with landscape orientation (Ncg = 6), while Fig. 7b) refers to
The adjacent lines also cast hard shadow on the PV modules in case portrait orientation (Ncg = 2). The diagrams clearly show the benefit of
of low solar elevation angles. For parallel support structure lines, the landscape orientation in reducing shading losses.
modules on a row are divided to a shaded and an unshaded part by a
line parallel to the rows. The wsh proportion of the shaded width of the 3.3. Photovoltaic module model
rows is calculated with the following equation developed using basic
geometric relations: The power of a PV module depends on the irradiance on the module
surface and the cell temperature. The cell temperature is affected by the
0 if cot S cos( S ) < cot
ambient temperature, the irradiance and the wind speed. Several
wsh = 1 if S 0
cot S cos( S ) cot
otherwise
cot S cos( S ) + cot (16)

A graphic explanation of the shading scene and notations is shown


in Fig. 5. The horizon shading caused by surrounding mountains is
calculated using a horizon profile. In case the Sun is under the horizon
line, the whole module surface is considered shaded.
The hard shading affects the beam and the circumsolar diffuse ir-
radiance. Considering all above mentioned effects and losses, the global
irradiance on the PV module surface is calculated using (17) for the Fig. 4. Arrangement of PV support structure line and the shading of diffuse
unshaded and (18) for the shaded part: radiation.

215
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

Fig. 5. Hard shading of support structure lines.

Mattei et al. (2006):


UPV (v ) = 26.6 + 2.3v (20)

The heat transfer depends on the wind speed at the height of the
modules, while the meteorological databases often contain wind speed
for a different height. The wind speed around the modules can be cal-
culated by the following equation (Huld and Gracia Amillo, 2015):
0.2
hmod
vmod = vane
hane (21)

where hmod is the average distance of the modules from the ground and
hane is height of the anemometer, while vmod and vane are the wind
speeds at these heights, respectively.
Fig. 6. Bypass diodes in a common 60-cell PV module.
Many models have been created to describe the performance of the
PV modules. Some of them, like the one presented by Huld et al. (2011),
different models have been created to calculate the cell temperature as using empirical equations with parameters derived by fitting of mea-
a function of these factors. Schwingshackl et al. (2013) compared 8 surement data. Another way of PV modelling is based on the single-
different models with experimental results and the model created by diode equivalent circuit (Fig. 8), derived from the physical principles of
Mattei et al. (2006) proved to be the most accurate for crystalline PV a solar cell.
modules. This model calculates the Tc cell temperature as follows: The four-parameter model is used in this calculation by neglecting
UPV (v ) Ta + GPV [ (1 µP , mp TSTC )] the effect of the shunt resistance. This simplified model is more favor-
c STC
Tc = able to use in optimization as it has lower computational costs than the
UPV (v ) + GPV STC µP , mp (19) five parameter model, while this simplification has only a minor effect
where Ta is the ambient temperature, αc is the absorptance of the solar on the precision of the MPP calculation (Dolara et al., 2015). The re-
cell, τ is the transmittance of the glass cover, ηSTC is the efficiency and sulting four-parameter model has the following equation:
μP,mp is the temperature coefficient of the module’s maximum power V + IRs
under standard test conditions (STC): GSTC = 1000 W/m2, I = IL I0 exp 1
a (22)
TSTC = 25 °C. The UPV heat transfer coefficient of the PV module is a
linear function of the v wind speed according to the parametrization of where I and V are the current and voltage of the module, IL is the light

Fig. 7. Shading loss factor as a function of the shaded area for landscape (a) and portrait (b) module orientations.

216
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

is almost unaffected. Correspondingly, the reduction of the MPP current


is calculated in the same way as the MPP power and the MPP voltage is
considered the same for each year. Han and Lee (2018) and Smith et al.
(2012) also highlighted that each PV modules degrade differently,
which causes increasing mismatch losses. Therefore, the degradation is
higher for arrays than for a single module, which must be considered
when choosing the appropriate degradation rate. The further calcula-
tions are performed individually for each year considering a different
Fig. 8. Single diode equivalent circuit of a solar cell. magnitude of degradation.

current, I0 is the diode reverse saturation current, Rs is the series re- 3.4. Cable, inverter and other electric losses
sistance and a is the modified diode ideality factor. The irradiance and
temperature dependence of these parameters are calculated according In photovoltaic systems Ns PV modules are connected in series in a
to De Soto et al. (2006), Dolara et al. (2015), Siddique et al. (2013): string and Np strings are connected in parallel to the input of the in-
a Tc verter. The PDC input power, VDC input voltage and IDC input current of
= an inverter is calculated as follows, considering the power loss and
aref Tc , STC (23)
voltage drop on the DC wires:
IL G
= PV [1 + µI , sc (Tc Tc , STC )] IDC = Np Imp (31)
IL, ref GSTC (24)
VDC = Ns Vmp Imp R c, DC (32)
3
I0 Tc Ns Tc, STC
= exp 1 PDC = Ns Np Pmp Imp2Rc, DC (33)
I0, ref Tc, STC qaref Tc (25)
In case of hard shading a modified version of these equations is
Rs = Rs, ref (26) used, taking into account the activation of bypass diodes as described in
where ε is the band gap of the semiconductor, ε = 1.121 eV for silicon the last section. The Rc,DC average resistance of DC string cables is
cell. calculated from the ΔvDC relative design voltage drop (at STC):
There are many different ways to determine the reference (STC) Ns Vmp, STC
values of these parameters, Humada et al. (2016) even composed a R c, DC = vDC
Imp, STC (34)
review on this topic. Three equations are typically derived from (22)
applied to open and short circuit and maximum power points: The efficiency of the inverter is calculated using the efficiency curve
provided by the manufacturer in the datasheet of the device, as a
IL, ref = Isc, STC (27)
function of the load at three input voltage levels. According to Bakhshi
et al. (2014), Rampinelli et al. (2014) the losses of the inverter can be
Uoc, STC
I0, ref = IL, ref exp approximated by a second-degree polynomial as a function of the in-
aref (28) verter power. The inverter efficiency is calculated by the following
equation as a function of the DC input power at a given V voltage level:

Rs, ref =
aref ln 1 ( Imp, STC
Isc, STC ) Ump, STC + Uoc, STC
K 0 (V)
inv (V) = + K1 (V) + K2 (V)(PDC / Pinv, max )
Imp, STC (29) (PDC / Pinv, max ) (35)
where Isc is the short circuit current, Voc is the open circuit voltage and where K0, K1, K2 are the coefficients determined by fitting the equation
Imp and Vmp are the current and voltage at maximum power point. to the datasheet efficiency curves using the least-squares method. Se-
Another equation can be derived from the temperature dependence of parate coefficients have been computed for all three voltage levels,
the module, but this method had proved to be sensitive to the tem- which are then used to calculate the power-dependent efficiency for
perature coefficients of the module published in modules’ datasheets these three voltages. The final power- and voltage-dependent efficiency
and often gave unreal results. A more robust method is implemented in is calculated with linear interpolation as a function of the DC voltage.
this work using the ηrel,200 relative module efficiency at 200 W/m2 ir- In case the input voltage is out of the operating voltage range of the
radiance, typically 96–97% compared to STC, specified in many PV inverter the device is considered to shut down and no power is pro-
module datasheets. The aref parameter is adjusted iteratively until the duced, although this happens rarely in properly sized PV systems. The
model represent the proper low-light performance. considered overload protection of the inverter limits the DC input
The maximum power point (MPP) of the I-V curves, resulting from power to the maximum input power of the inverter by regulating the
(22) for each irradiance and temperature, is obtained with a golden DC voltage from MPP to another point with lower module power (Chen
ratio search algorithm. The model provides not only the Pmp maximum et al., 2013). The PAC output power of the inverter is calculated with the
power, but also the Imp current and Vmp voltage at MPP. following equation, considering both the ηinv inverter efficiency and the
The reduction of the module performance due to degradation is also clipping of high input power:
considered. One type of degradation is light-induced degradation (LID),
which occurs in the first hours of the exposure to the solar radiation
PAC = min( inv (PDC , VDC ) PDC , Pinv, max ) (36)
(Lindroos and Savin, 2016), while the other type is a continuous de- Ground-mounted PV plants typically connect to the grid on medium
gradation over the whole lifetime. These effects are calculated using lLID or high voltage, therefore step-up transformers are used to provide the
loss factor and ldeg yearly degradation rate as a decrease of the module required voltage level (Cabrera-Tobar et al., 2016). The total power on
power for the tth year of the lifetime: the low voltage side of the transformers is calculated with the following
equation, taking into account the losses on the low voltage AC cabling
ldeg )t (30)
Pmp, t = Pmp,0 (1 lLID )(1 0.5
assuming no reactive power (cosφ = 1):
According to the long-term measurement results of 12 crystalline 2
silicon PV modules it is concluded by Smith et al. (2012) that the de- PAC
Ptr = Ninv PAC Rc , AC
gradation mainly reduce the current of the modules while their voltage Vl (37)

217
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

where Ninv is the number of inverters in the GCPV plant. The Rc,AC
average resistance of AC cables from the inverters to the transformers is
derived from the ΔvAC relative design voltage drop at nominal inverter
power:
Vl 2
R c, AC = vAC
Pinv, max (38)

Medium voltage grid connection is typically allowed for plants up to


5–20 MW nominal power, depending on the local regulations and grid
topology (Malamaki and Demoulias, 2014). The low to medium voltage
(LV/MV) transformers are placed among the PV rows in the proximity
of the inverters to minimize the low voltage AC cable distances. In case
a high voltage connection is imposed by the distribution systems op-
erator (DSO), one or several medium to high voltage (MV/HV) trans-
formers or even a substation has to be installed. New transmission lines
may also be required between the site and the grid connection point. As
the number, length and even the presence of the transformers, medium
and high voltage cables and switchgears largely depend on the local
conditions, they are not modeled in detail. The losses of these AC
Fig. 9. Schematic of the support structure.
components are considered using an ltr transformer loss factor to cal-
culate the Pgrid power fed into the grid:
Nmod
Lst = Lmod
Pgrid = Ptr (1 ltr ) (39) Nmpl (46)
The structure consists of repeating segments of Lseg length along the
4. Economic model total length of the lines. The arrangement of one segment is visualized
in Fig. 9.
The economic model estimates the initial installation as a function Each segment consists of one tilted spar with a length of W, one
of the optimized design parameters, estimates the yearly operation and vertical spar for every Wstand width of the structure and horizontal spars
maintenance (O&M) costs then using them along with the annual rev- on the two sides of the structure and between each module rows. Pile
enues to calculate the profitability of the GCPV plant. foundation is considered, which requires H0 extra length for each ver-
tical spar measured from the lowest part of the modules according to
4.1. Installation costs Fig. 9. The Bvert total length of vertical spars, using these assumptions, is
calculated as follows:
The CPV cost of the PV modules is calculated based on the cmod price W W sin
of one module and the total number of modules (Sulaiman et al., 2012): B vert = ceil H0 +
Wstand 2 (47)
CPV = cmod Nmod = cmod Ns Np Ninv (40)
The Btot total equivalent length of the support structure spars, con-
Similarly, the Cinv cost of the inverters is computed using the cinv sidering the cross-section of the horizontal spars as the half of the tilted
price of one inverter and the total number of inverters: and vertical ones, is calculated with the following equations:

Cinv = cinv Ninv (41) Lst Lseg (Nmpl + 1)


Btot = B vert + W +
The Cland cost of the occupied land area, including the land levelling
Lseg 2 (48)
works, the lighting and lightning protection and other costs depending The Cc,DC cost of DC cabling is calculated using the length of DC
on the land area is proportional to the Aland land area and the cland unit cables. In case of string inverters, considered in this study, the strings
cost (Aronescu and Appelbaum, 2017): connecting to the same inverter are typically placed in the same support
Cland = cland Aland (42) structure line to avoid underground cable ducts. According to the op-
timal layout presented in Fig. 10, the Lc,DC average length of the string
The land area consists of an Al,0 constant term for the main control DC cables is calculated as follows:
building and the Aland,mod land area occupied by the modules on the
Ns Np Lc 0, DC
support structure lines (according to Fig. 4), multiplied by an aland Lc , DC = Lc 0, DC +
factor accounting for the extra land area of the inverters, transformers, Nmpl (49)
cable channels, roads, ditches, etc: where Lc0,DC is the basic length of the cables, accounting for vertical
Aland = Al,0 + aland Al, mod (43) height differences and margins. This length only covers the red cables
of Fig. 10 since the modules are connected to each other by their own
sin cables.
Al, mod = Nmod Lmod Wmod cos +
tan (44) The Ac,DC cross-section of the cable is calculated using its length and
resistance (Gómez-Lorente et al. (2014):
where Wmod and Lmod are the width and length of a PV module.
The Cstruct cost of the support structures are considered proportional Cu Lc , DC
Ac , DC =
to the raw material, i.e. the Btot total length of the steel spars according Rc , DC (50)
to Kornelakis and Marinakis (2010): where ρCu is the resistivity of copper. The total cost of DC cabling is the
Cstruct = cstruct Btot (45) product of the total length of DC cables and the cc,DC(Ac,DC) cross-sec-
tion dependent unit cost of the cables:
The total length of the support structure lines is calculated as fol-
lows: Cc, DC = cc , DC (Ac, DC ) Ninv Np Lc, DC (51)

218
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

Fig. 10. String layout for DC cable length cal-


culation.

The Cc,AC cost of low voltage AC cabling is calculated similarly to costs like planning, transportation, installation and taxes. They can be
the DC cables. The power of one transformer is considered the same as either constant or any arbitrary function of any of the design or cal-
the total nominal power of the inverters connecting to it. In case the culated parameters, which provides a great flexibility in refining the
cables run only parallel or perpendicular to the support structure lines, economic model.
the optimal layout of the inverters is a square around the transformer The Cinst total installation cost of the GCPV plant is the sum of the
according to Fig. 11 (Aronescu and Appelbaum, 2017). Using this as- described components:
sumption, the Lc,AC average length of an AC cable is calculated as fol-
Cinst = CPV + Cinv + Cland + Cstruct + Cc, DC + Cc, AC + Ctr (56)
lows:

1 P1tr
Lc , AC = Lc 0, AC + Al, mod 4.2. Payback calculation
2 Pnom, AC (52)

where P1tr is the nominal power of one LV/MV transformer and Pnom,AC The Rt annual revenues of the GCPV plant in the tth year, coming
is the total nominal AC power. from the sale of the produced electricity, is calculated as the sum of the
The Ac,AC cross-section of one conductor of the three-phase AC cable product of the pel unit rate of electricity, the Pgrid,t,i power fed into the
is calculated using its length and resistance: grid in each timestep of the given year and the Δt length of a timestep:

Cu Lc , AC Rt = pel Pgrid, t , i t
Ac , AC = (57)
Rc , AC (53) i

The electricity rate includes all subsidies and it can be either con-
The total cost of AC cabling is the product of the total length of AC
stant or varying during and between years, which makes it possible to
cables and the cc,AC(Ac,AC) cross-section dependent unit cost of the
assess the effect of prices changes during the lifetime of the plant.
cables:
The CO&M,t yearly operation and maintenance (O&M) cost are as-
Cc, AC = cc , AC (Ac, AC ) Ninv Lc, AC (54) sumed proportional of the initial installation costs with an αO&M cost
factor:
The Ctr cost of the AC components, like transformers and medium
voltage (MV) cables and switchgears is considered proportional to the CO & M , t = O & M Cinst (58)
Pnom,AC total nominal AC power:
The probability of failure of the inverters is considered uniform over
Ctr = ctr Pnom, AC = ctr Ninv Pinv (55) time, thus the Crepl,t inverter replacement cost is the same in each year:

The cmod, cinv, cland, cstruct, cc,DC, cc,AC and ctr unit cost factors, in Cinv
Crepl, t =
addition to the purchase costs of the devices, also includes all additional MTBFinv (59)

Fig. 11. Inverter layout for AC cable length calculation.

219
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

where MTBFinv is the mean time between failures of the inverter minimum so all the parameters can have any possible, even unrealistic
(Bakhshi and Sadeh, 2016). values. This attitude helps to identify any possible error in the para-
The upfront incentives granted for the GCPV plant are typically metrization of the model as an appearance of any unreal values in the
proportional to the total installation cost or the nominal DC or AC optimization results is a sign of an error.
power of the plant (Chen et al., 2013). These S subsidies can be cal- Due to the complexity of the model, the IRR is not a unimodal
culated as an arbitrary function of these parameters: function of the design parameters, therefore the gradient-based opti-
mization method cannot guarantee to find the global optimum. A ge-
S = f (Cinst , Pnom, DC , Pnom, AC ) (60)
netic algorithm (GA), one of the most widespread metaheuristic evo-
The profitability is calculated using the net present value (NPV) lutionary algorithms, is used for the optimization. The model
method (Gómez et al., 2010): calculations implemented in MATLAB; therefore, the built-in genetic
tlt algorithm of MATLAB is used for the optimization.
Rt CO & M , t Crepl, t
NPV = (1 rtax ) (Cinst S) The genetic algorithm starts by generating an initial population of a
t=1
(1 + r )t (61) number of individuals defined by the population size property. The set
of parameters, also called chromosomes, of the individuals are chosen
where tlt is the design lifetime of the plant, rtax is the taxation rate on the
randomly with a uniform distribution between the boundaries of each
net incomes and r is the discount rate.
decision variables. The IRR objective function is calculated for each
The internal rate of return (IRR) is calculated as the discount rate
individual using the presented mathematical models to determine their
where the NPV is zero:
raw fitness scores, which are then converted to expectation values with
NPV (r = IRR) = 0 (62) a scaling function. The parents of the next generation are chosen based
on their expectation values. In an elitist genetic algorithm, a number of
This equation is solved numerically for the IRR.
individuals with the best fitness function directly survive to the next
generation. The rest of the next generation, knows as children, are
5. Optimization method
created either by crossover or mutation of the selected parents. The
proportion of crossover children is defined by the crossover fraction
The goal of the optimization is to maximize the internal rate of
property and they are created by combining the chromosome vectors of
return (IRR) objective function. The ten decision variables with their
two parents. The rest of the children are created by the mutation of one
boundaries and constraints are listed in Table 1.
parent. After the creation of the next generation the process is starting
The manufacturers sort the PV modules into categories based on
over from the evaluation of the objective functions until a stopping
their measured power at STC conditions with 5 W increments. The Pmod
criterion is met. The termination condition is that the average relative
module power is an integer divisible by 5 with a Pm,min minimum and
change in the best fitness function over a number of stall generations is
Pm,max maximum values depending on the supply of the chosen module
lower than the function tolerance property. The process of the genetic
type. The Ns number of series connected modules in a string is also an
algorithm is summarized in Fig. 12.
integer. The total open circuit voltage of the PV modules must not ex-
A more detailed description of the algorithm and its parameters can
ceed the maximum system voltage limited by either the module or the
be found in the documentation of the software (MathWorks, 2018). The
inverter even at the lowest design temperature, which constrains the
chosen values of the optimization parameters are listed in Table 2. The
Ns,max maximum number of series modules:
shift linear scaling function sets the expectation value of the best in-
min(Vmax , mod, Vmax, inv ) dividual 2 time higher than the average score and scales all the others
Ns, max =
Voc, STC [1 (Tc, STC Tc, min ) µV , oc ] (63) linearly. The scattered crossover function choses each parameters of the
child randomly either from the first or second parent. The gaussian
In practical PV system design, similar constraints are also for- mutation function adds a random number taken from a gaussian dis-
mulated for the MPP voltage to ensure that the string voltage is always tribution to each parameter of one parent. The mean of the distribution
in the operating voltage range of the inverter. In this study, however, is 0 and the standard deviation reduces continuously with the in-
the inverter model calculates the real power losses in case of improper creasing number of generations to avoid big changes when a proximity
input voltage, and this penalty makes it unnecessary to define a of the global optimum has already been reached.
minimum number of series modules as a hard constraint. The Np Two important changes from the default settings are (i): increase of
number of parallel strings connecting to an inverter is an integer and its the population size and (ii): increase of the mutation fraction by de-
Np,max maximum value is limited either by the number of inputs or the creasing the crossover fraction, both aiming to enhance the variability
maximum input current of the chosen inverter. If no maximum number during the optimization. Without these modifications the algorithm has
of string inputs is specified on the inverter datasheet, Np,max is calcu-
lated by dividing the maximum DC input current by the STC current of
the PV modules. Table 1
The Ninv number of inverters is constrained by a Ninv,min minimum Decision variables of the optimization with boundaries and constraints.
and Ninv,max maximum value, standing for an expected nominal power Name Dim. Min Max Const. Description
range. The Nmpl number of modules along the width of a support
structure line is positive integers, but typically not higher than 8, Pmod Wp Pm,min Pm,max 5∙INT STC power of a PV module
Ns 1 1 Ns,max INT Number of modules connected series
therefore a limit of 20 is quite permissive. The ΔvDC and ΔvAC voltage
in a string
drops are continuous variables ranging from 0 to 1. The β tilt and γ Np 1 1 Np,max INT Number of parallel strings connected
azimuth angles are continuous and ranging from 5 to 90° and −180° to to an inverter
180°, respectively. The α angle, representing the relative inter-row Ninv 1 Ninv,min Ninv,max INT Total number of inverters
spacing is continuous, positive and has an αmax upper boundary re- Nmpl 1 1 20 INT Number of module rows in one
mounting structure line
presenting a Dmin minimum inter-row spacing required for proper in-
ΔvDC 1 0 1 Relative voltage drop on the DC
stallation and maintenance: cables
ΔvAC 1 0 1 Relative voltage drop on the AC cables
1
Nmpl Wmodsin β ° 5° 90° Tilt angle of PV modules
max = tan
Dmin (64) γ ° −180° 180° Azimuth angle of PV modules
α ° 0° αmax Angle representing inter-row spacing
The boundaries and constraints represent the bare physical

220
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

year datasets while still provide acceptable accuracy for the demon-
stration of the optimization method. The TMY data and the horizon
profile were downloaded for Budapest (47.498° N, 19.040° E) based on
the data of 2007–2016 from PVGIS5 (EC Joint Research Centre, 2017).

6.1. Optimal plant configuration in Hungary

The optimization was run 100 times and Table 5 summarizes the
best result and the mean and standard deviation of all results for the
Fig. 12. Flowchart of the genetic algorithm. objective function and the 10 decision variables. The results are co-
herent as the variance of the parameters and the IRR is almost negli-
Table 2 gible. The optimal values of the decision variables are realistic as they
Parameters of the genetic algorithm. are in the range of real values used in GCPV plants. The number of
Population size 100 modules in series and parallel and the power of a module results is a
Scaling function Shift linear AC/DC power ratio of 67.48% for the best configuration, which is even
Elite count 5 lower than the 73.53% ratio introduced by Fu et al. (2018) as a typical
Crossover fraction 70% value. The plane of maximum irradiation has 34.81° tilt angle and
Crossover function Scattered
−7.09° azimuth based on the used meteorological database. The op-
Mutation function Gaussian
Function tolerance 10−6 timal tilt and azimuth angles are lower mostly due to the shading effect,
Stall generations 50 which corresponds to the results of previous studies (Kornelakis, 2010;
Vokony et al., 2018). The slightly southeast orientation represented by
the negative azimuths is due to the higher irradiation in the morning
stuck in local maximums with an unacceptably high probability. The hours than in the afternoon in the TMY database, which highlights the
scaling and selection functions are also changed as the chosen functions importance of the accuracy of the used meteorological database. The
ensured faster convergence for the algorithm. α = 17.9° angle means that no shading occurs on the day of winter
A reasonably close proximity of the global optimum was found in all solstice between 1:14 h before and after the solar noon, which is close
cases of 100 runs of the algorithm for the input data presented in the to the typical rule of thumb for row spacing calculation. The nominal
next section. The average computation time was 396.9 s on a laptop voltage drops are both close to the typical 1% design value with a
with an Intel i7-7700HQ 2.80 GHz processor and 32 GB RAM. The time slightly higher value on the AC side due to the longer length of cables.
and computation requirements are acceptably short for a wide range of An investment is considered economically feasible if the IRR is
possible applications of the method. higher than the weighted average cost of capital (WACC) of the in-
vesting company, although the risk is also essential in the economic
evaluation. Due to the low interest rate environment prevailing in in-
6. Results and discussion ternational markets today, even this IRR of around 8% is feasible,
especially as the electricity is sold in a risk-reducing, predictable feed-in
This section describes a demonstration of the optimization method tariff system. The numerous PV power plants installed in Hungary in
for a plant site near Budapest, Hungary. The components and technical previous years, which mostly have suboptimal layout and therefore
parameters considered for the optimization are detailed in Table 3, lower profitability, also demonstrates that this rate of return is accep-
while the economic parameters and cost functions are listed in Table 4. table by most of the investors.
All the parameters are based on Hungarian regulations and real costs Table 6 contains the maximum irradiance gain resulting from the
and technical data of several GCPV plants installed in Hungary in the tilted module placement and all the important losses of the energy
last 2 years with nameplate capacities in between 0.5 and 20 MW. The conversion taken into account in the performance model. The specific
minimum and the maximum number of inverters are set to 25 and 250, energy production of the plant is 1186 kWh/kWp for the first year and
respectively, representing a nominal power range from 1 to 10 MW. In 1135 kWh/kWp on average over the whole lifetime. The levelized cost
this range, the plant connects to the grid at medium voltage, and the of electricity (LCOE), calculated with a 7% discount rate, is 92.8
loss and cost factors of the transformers are chosen accordingly. €/MWh.
The temporal resolution of the meteorological database is important
for the accuracy of the optimization. High resolution makes the results Table 3
more reliable but increases the computation time as more calculation is Components and technical parameters of the GCPV plant.
required to obtain the annual performance. On the other hand, long
Module: Canadian Solar CS6K- 285|290|295|300|305P
timesteps introduce error by vanishing short-time effects like shading Pm,min 285 W Pm,max 305 W
around sunrise and sunset. Many design software tools, like PVGIS and ηrel,200 96.5% Vmax,mod 1500 V
PVsyst, and many research papers, like (Bakhshi and Sadeh, 2016;
Kornelakis and Marinakis, 2010; Notton et al., 2010), are using me- Inverter: Huawei SUN2000-36KTL
teorological database with hourly timesteps. This was implemented in Pinv,max 40 kW K0(580 V) −0,00186
this sample calculation as it offers a good tradeoff between time and Vmax,inv 1100 V K1(580 V) 0,98766
Np,max 8 K2(580 V) −0,00537
performance. The 5th version of the Photovoltaic Geographical In-
K0(720 V) −0,00134 K0(850 V) −0,00205
formation System (PVGIS5) by European Commission offers a freely K1(720 V) 0,99328 K1(850 V) 0,99147
available Typical Meteorological Year (TMY) dataset for most of the K2(720 V) −0,00601 K2(850 V) −0,00728
world. The TMY dataset contains the hourly data of a typical year by
selecting the most representative data for each month from a measured lLID 2% Dmin 3m
10-year-long database, which reduces the uncertainty resulting from ldeg 0.5%/a Al,0 100 m2
the year-to-year variance of the weather compared to the data of one ltr 2% L0,cDC 8m
Wstand 2m L0,cAC 10 m
arbitrarily chosen year. Even though the TMY databases are not able to
Lseg 3m P1tr 1 MW
capture the interannual variability of solar resources (Bryce et al., 2018; H0 2.5 m MTBFinv 12 a
Vignola et al., 2012), they ensure faster calculation compared to multi-

221
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

Table 4
Economic parameters of the optimization.
cmod 130 + 0.8∙(Pmod – Pm,min) €
cinv 4925 €
cland 3.6 + 0.06∙Ac,DC[km2] €/m2
cstruct 20.2 €/m
cc,DC 1.2 + 0.12∙Ac,DC[mm2] €/m
cc,AC 8 + 0.35∙Ac,AC[mm2] €/m
ctr 135 €/kW
pel 0.098 €/kWh
rtax 5%
tlt 25 a
αO&M 2%
Fig. 13. Installation cost structure of the optimal GCPV plant configuration.

Table 5
Average and standard deviation of decision variables and objective function
after 100 GA optimization runs.
Name Best Average Standard deviation

Pmod 285 W 285 W 0W


Ns 25 25 0
Np 8 8 0
Ninv 114 115.9 4.3
Nmpl 6 6.5 0.9
ΔvDC 0.66% 0.63% 0.064%
ΔvAC 1.44% 1.44% 0.014%
β 19.7° 19.1° 0.8°
γ −2.0° −2.2° 0.3°
α 17.9° 17.6° 0.5°
Fig. 14. Average monthly energy generation over the lifetime.
IRR 8.231% 8.229% 0.0030%

Table 6
Losses of the optimal GCPV plant configuration.
Plane of maximum irradiation (gain over horizontal) +14.42%
Real module orientation −2.50%
Reflection −3.17%
Diffuse shading −1.47%
Soiling −1.09%
Photovoltaic conversion at STC efficiency −82.59%
High temperature efficiency decrease −1.83%
Low irradiance efficiency decrease −1.18%
Hard shading −0.49%
Light induced degradation −2.00%
Average degradation −6.01%
DC cables −0.37%
Inverter overload −2.68%
Inverter voltage limits −0.00%
Inverter conversion −1.65% Fig. 15. Annual cash flow and NPV with 6% and 8% discount rates.
AC cables −1.01%
Transformer and medium voltage −2.00%
the discount rate, the payback time reduces to 17 years. This value is
rather favorable, considering that the profitability is calculated with a
relatively low feed-in tariff and without any other subsidies.
Fig. 13 shows the distribution of the GCPV plant installation cost
among the main categories calculated by the economic model. The
specific investment cost is 878.9 €/kWp. The cost structure corresponds 6.2. Sensitivity analysis
to the typical results presented in Fu et al. (2018), IRENA (2018).
The average monthly energy generation diagram of the PV plant is A sensitivity analysis was carried out to quantify the effect of the
shown in Fig. 14. The distribution of the generated electricity and the non-optimal design parameters. The analysis was performed ceteris
large difference between the production during winter and summer is paribus, so only one decision variable was changed at a time while the
mostly due to the local climate and the availability of solar irradiance, others had their optimal values.The results of the analysis are visible in
although, to a smaller extent, the design parameters also affect this Figs. 16 and 17.
diagram. The module power has only a minor effect on the profitability since
Fig. 15 visualize the annual cash flow and the NPV over the design manufacturers price all their modules according to their real value to be
lifetime of the PV plant. After the substantial initial investment cost in able to sell all of them. The number of series and parallel modules has
year 0, there are continuously positive but gradually decreasing cash the highest effect on the IRR as they directly affect the AC/DC power
flows due to the degradation of the PV efficiency. The economic pay- ratio, i.e. the inverter sizing ratio, which is the most common and sig-
back time is the year when the NPV of the project becomes positive, nificant optimization variable in PV systems (Khatib et al., 2013). The
which happens only in the last year of operations for a discount rate of effect of the number of inverters, reflecting the size of the PV plant is
8%. However, if the cost of capital is low enough for 6% to be used as small since only few size-related cost functions are included in the
economic model. The number of module rows favorize even number

222
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

Fig. 16. The sensitivity of the IRR to the five discrete decision variables.

due to the definition of the mounting structure and has acceptably good annual energy production (lifetime average), the C0/PDC specific in-
IRR for the common 4, 6 and 8 rows. vestment cost and the LCOE.
The design voltage drops have a slight effect on the IRR since both According to the results, the location has a significant influence on
the losses and costs associated with cabling is relatively small. The the optimal value of most PV plant design parameters. In terms of the
module orientation and inter-row spacing are important design para- inverter sizing, the AC/DC ratio is significantly lower in two locations
meters as they all have a high effect on the profitability. The sensitivity (Singapore and London) compared to the others. The common feature
of the IRR to the continuous design variables is approximately quad- in the meteorology of these two locations is the high diffuse fraction
ratic. Consequently, the small deviations from the optimal layout has (52% in Singapore and 57% in London), which indicates a more cloudy
almost negligible effect but a higher difference can significantly de- climate resulting in a lower number of sunny days and less frequent
crease the profitability. near-peak power of the PV modules. In these circumstances, a lower
inverter power is optimal as it only slightly increases the overload losses
while it benefits from a better part-load efficiency on the cloudy days.
6.3. Effects of location and meteorology The lower DC/AC power in the other locations is achieved by using a
lower number of parallel strings (Np), as this also reduces the DC cables
The performance of a PV plant highly affected by the geographical losses and costs, and therefore it is more beneficial than decreasing the
location and meteorology of the site, therefore it is important to identify number of series modules in the strings (Ns).
the effect of these factors on the optimal design. The location has a The optimal design values of DC and AC cable losses are increasing
direct effect through the path of the Sun and the incidence angle of with the decrease of the global irradiation of the locations. Lower ir-
beam radiation, while it also determines the meteorological char- radiance results in a lower power on average, and due to the quadratic
acteristics and the available solar resource. In this section, optimal dependence of the cable losses on the electric current, it also leads to a
plant configuration is presented for five different locations. The results smaller fraction of the yearly average cable losses compared to their
are based on the TMY datasets downloaded from PVGIS5. In practice, design values. In more insolated areas thicker cables should be used to
the cost functions and especially the feed-in tariff may also change maintain the same annual cable loss rates compared to areas with less
between the locations. However, including these cost changes would solar resources.
mix the effects of location with the effects of economic parameters; The tilt angles generally depend on the latitude, as for higher lati-
therefore the same costs, listed in Table 4, are used for all locations. tudes higher tilt is necessary for the best incidence angle of the beam
The five cities were chosen from significantly different parts and radiation. The only exception is London, where the algorithm resulted
climate zones around the world. The optimization results are presented in only a 13.9° tilt angle despite being the northernmost location with
in Table 7. In each row, the name of the location is followed by the 51.51° latitude. This is caused by the high diffuse fraction of the loca-
geographic coordinates and three important characteristics of the local tion as a lower tilt is more beneficial to collect the most of the isotropic
meteorology, then Ta,mean yearly mean temperature, the ΣGh annual sky diffuse radiation. According to the results, the optimal tilt angle in
sum of global horizontal and ΣDh diffuse horizontal irradiation, all of ground-mounted PV plants is significantly lower compared to the rule
them calculated directly from the TMY database. They are followed by of thumb suggesting to choose the tilt angle equal to the latitude, which
the optimal values of the 10 design variables, the IRR objective func- is a common consequence of the row shading, the diffuse fraction and
tion, the PAC/PDC AC/DC nominal power ratio, the Eyr/PDC specific

Fig. 17. The sensitivity of the IRR to the five continuous decision variables.

223
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

Table 7 6.4. Effects of PV module costs


Optimization results for different locations.
Location Dubai Johannesburg Singapore Boston London Beyond the location, meteorology and technical aspects, the costs of
the different components also influence the optimal design. The most
Coordinates 25.06° N, 26.21° S, 1.34° N, 42.37° N, 51.51° N, determining components are the PV modules, covering around half of
55.50° E 28.05° E 103.83° E 71.05° W 0.13° W
the total installation costs (see Fig. 13). The price of PV modules has
Ta,mean, °C 28.3 15.3 27.3 10.9 10.2
ΣGh, kWh/m2 2305 2054 1723 1462 996
dropped by more than 80% in the last decade, which decrease is ex-
ΣDh, kWh/m2 742 656 901 536 567 pected to continue even in the following years (IRENA, 2018). There-
fore, it is important to explore the decreasing module prices’ effects not
Pmod, W 290 285 285 285 290 only on profitability and LCOE but also on the optimal plant design
Ns 26 26 26 26 26 configuration.
Np 7 7 8 7 8 The optimization was performed for different PV prices ranging
Ninv 176 171 166 127 119
from 120% to 50% of the original cmod module cost function introduced
Nmpl 8 8 8 6 8
ΔvDC 0.35% 0.40% 0.49% 0.52% 0.53% in Table 4. The costs of the other components were considered constant
ΔvAC 1.20% 1.25% 1.24% 1.45% 1.55% as their variation is much slower compared to the PV modules. The
β, ° 11.8 14.5 5.0 22.9 13.9 results for Budapest are summarized in Table 8.
γ, ° 3.5 −177.4 −147.7 −0.4 −6.0 A 50% drop of the PV costs results in a 25% reduction of the LCOE
α, ° 24.0 23.6 29.9 19.5 14.0
of an optimal plant. Moreover, the main effects of the PV module cost
reduction are the following:
IRR 16.72% 15.41% 11.29% 10.09% 4.25%

PAC/PDC 75.8% 77.1% 67.5% 77.1% 66.3%


• The AC/DC ratio is decreasing, which increases the inverter over-
load losses but reduces the relative costs of the inverter and other
Eyr/PDC, kWh/ 1823 1731 1319 1323 853
kWp AC components.
C0/PDC, €/kW 880 887 845 908 870 • The nominal cable voltage drops are increasing, which increase the
LCOE, €/MWh 58.2 61.9 77.0 82.7 121.9 cable losses but reduce their size and costs.
• The tilt angle is decreasing, which reduces the in-plane irradiation
and consequently the energy production of the PV modules, but also
the higher irradiation in summer than in winter. The optimal azimuths contributes to the decrease of the height and cost of mounting
are close to 0° and −180°, the values representing that the surface is structures, the required land area and cable lengths.
facing the equator, the slight difference is due to local effects. The ex-
ception is Singapore, which is so close to the equator and the modules
• The row spacing is decreasing, which increases the shading losses
but reduces the land area and cable lengths.
have such a low tilt angle that the surface azimuth has only a minor
effect and local factors determine its optimum. The inter-row spacing
• The number of module rows per mounting structure line is in-
creasing, as the more densely packed PV modules help to decrease
angle is decreasing with moving farther from the equator as a bigger DC cable lengths and shading losses, while the increase in the
row distance is necessary to avoid increased shading for the lower solar mounting structure cost is not significant due to the low tilt angle.
elevation angles.
In terms of IRR, the locations with higher irradiation seems much
• The number of inverters is increasing as the decreased relative land
usage enables to install more power to roughly the same area.
more favorable for PV investments; however, this is not always true
from the standpoint of the investor. The IRR was calculated with the As a summary, all these changes aim to decrease the installation
same feed-in tariff for all locations, while in reality, the feed-in tariff is costs of other components at the price of increased system losses and
typically aligned to the solar resource to ensure a fair, but not excessive decreased specific energy production (see the Eyr/PDC of Table 8). The
financial return. Among the results of Table 7 the LCOE gives a better decreasing AC/DC ratio is also beneficial for the grid due to its peak
indication of the cost of solar electricity production. shaving effect, while the reduced land usage helps to install a PV plant
of higher capacity even in limited areas. Both of these favorable ten-
dencies will be strengthened by decreasing PV costs in the future.

Table 8
Optimal PV plant design for different module costs.
PV module cost 120% 110% 100% 90% 80% 70% 60% 50%

Pmod, W 285 285 285 285 285 300 305 305


Ns 25 26 26 26 26 25 25 26
Np 8 8 8 8 8 8 8 8
Ninv 110 112 116 119 124 131 137 143
Nmpl 6 6 6 6 8 8 8 8
ΔvDC 0.58% 0.65% 0.66% 0.68% 0.55% 0.56% 0.60% 0.65%
ΔvAC 1.42% 1.43% 1.44% 1.46% 1.48% 1.48% 1.50% 1.51%
β, ° 21.5 19.8 19.7 19.1 15.9 15.2 13.9 11.3
γ, ° −2.2 −2.4 −2.0 −2.0 −2.3 −2.1 −2.6 −2.6
α, ° 16.5 16.6 17.9 18.0 17.7 17.9 18.0 18.1

IRR 6.84% 7.51% 8.23% 9.02% 9.88% 10.85% 11.98% 13.27%

PAC/PDC 70.2% 67.5% 67.5% 67.5% 67.5% 66.7% 65.6% 63.1%


Eyr/PDC, kWh/kWp 1154 1139 1135 1133 1124 1119 1109 1086
C0/PDC, €/kW 984.4 927.3 878.9 831.6 779.4 730.0 675.9 614.6
LCOE, €/MWh 101.7 97.3 92.8 88.3 83.8 79.1 74.3 69.3

224
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

6.5. Applications and boundaries of the method plants could be simultaneously optimized.
The constructed PV performance and economic models offer a good
The method was developed and tested for a ground-mounted GCPV tradeoff between accuracy and computational requirements, making
plant in Hungary, however, with some modifications, the models can be them suitable for the optimization task. The proposed optimization
also adjusted to describe different systems. For example, roof-mounted algorithm provides coherent and reliable results with a calculation time
systems are another important category of grid-connected photo- of 6–7 min. The models can be easily tailored to any practical GCPV
voltaics. The household-scale PV systems are typically installed on a plant design task simply by changing the meteorological database, the
sloped roof with fixed orientation without structure rows and with short component properties and the cost functions. The relatively short cal-
cable lengths, therefore most of the presented design parameters and culation time enables to run multiple simulations with many different
either not applicable or irrelevant in this case. Industrial-scale rooftop components and parameters in order to support the decision making in
systems, typically installed on large flat roofs, have a similar layout to the system design. The commercialization of the presented method
the ground-mounted systems. Therefore, the presented methodology is could facilitate the optimal design of the GCPV plants and reduce the
able to model and optimize them with some modifications, like ac- costs associated with the design process.
counting for the unusable parts and the load capacity of the roof or the A possible research application of the method is to study the effect
electricity demand of the building. Other developing fields are plants of the geographical location or the cost changes of different components
with bifacial modules and floating PV systems. These plants could also on the optimal power plant configuration. The method could also be
be optimized with the proposed method, although, among adding new useful in the evaluation of new components and technologies in PV
equations to consider bifaciality, a more detailed calculation of re- plants by enabling to study an optimal power plant without consider-
flected irradiance is also necessary to maintain the reliability of the able efforts.
results. The further development of the presented method covers the as-
Technically the presented optimization method can be applied to sessment of different optimization algorithms in order to find the most
any location around the world by selecting a proper meteorological appropriate solution for this optimization problem. An improved al-
database for the given location, however, some adjustments might be gorithm or the hybridization with other heuristic methods could enable
necessary to account for the local features. In deserted areas soiling to include more decision variables, e.g. the selection and optimization
losses are a more important issue causing high losses, therefore their of the main components from a component database. Another devel-
more detailed modeling is necessary. The meteorological datasets are opment possibility is to include the environmental effect of the design
mostly based on historical data, which may be insufficient to describe variables and extend the optimization with an environmental objective
recently emerging factors like the effect of increasing air pollution or function to support the eco-design of the PV plants.
climate change on the solar resource in some areas.
The performance model calculates the hourly electricity production Declaration of Competing Interest
for the whole lifetime of the power plant, therefore either fixed or time-
dependent electricity prices can be used to calculate the revenues. In The authors declare that they have no known competing financial
this way, fixed feed-in tariff, feed-in premium or unsubsidized whole- interests or personal relationships that could have appeared to influ-
sale electricity market prices can all be used for the simulation de- ence the work reported in this paper.
pending on the incentives granted for the investment. In some countries
– even in Hungary for the new installations - PV plants have to purchase Acknowledgment
balancing energy for the deviation of their real and scheduled pro-
duction caused by the inaccuracy of weather forecasts, which leads to This work was supported by the National Research, Development
additional costs and reduce the financial return. These costs were not and Innovation Fund (NKFIH) of Hungary in the frame of FIEK 16-1-
yet implemented in the model due to the high uncertainty of the fore- 2016-0007 (Higher Education and Industrial Cooperation Centre) pro-
cast errors and balancing services cost. An increasingly common way to ject and by the National Research, Development and Innovation Fund
reduce balancing energy needs or shift production to the hours of high (TUDFO/51757/2019-ITM) Thematic Excellence Program.
demand is the installation of energy storage. The inclusion of storage in
performance modeling would highly complicate the model by in- References
troducing many new aspects, like charging and discharging strategies.
The calculation with storage and balancing energy are important areas Abdeen, E., Orabi, M., Hasaneen, E.S., 2017. Optimum tilt angle for photovoltaic system
of the further development of the method. in desert environment. Sol. Energy 155, 267–280. https://doi.org/10.1016/j.solener.
The cost of modules and inverters have largely decreased in the 2017.06.031.
Aronescu, A., Appelbaum, J., 2017. Design optimization of photovoltaic solar fields-in-
previous decade, therefore balance of the system (BOS) and soft costs sight and methodology. Renew. Sustain. Energy Rev. 76, 882–893. https://doi.org/
cover an increasing share of the total investment costs, especially in 10.1016/j.rser.2017.03.079.
developed countries where labor costs are higher (Fu et al., 2018; Bakhshi-Jafarabadi, R., Sadeh, J., Soheili, A., 2019. Global optimum economic designing
of grid-connected photovoltaic systems with multiple inverters using binary linear
IRENA, 2018). As the module and inverter costs are similar due to the programming. Sol. Energy 183, 842–850. https://doi.org/10.1016/j.solener.2019.
fierce competition, BOS and soft cost became the most important for the 03.019.
profitability of PV plants. The presented method offers an efficient way Bakhshi, R., Sadeh, J., 2016. A comprehensive economic analysis method for selecting the
PV array structure in grid-connected photovoltaic systems. Renew. Energy 94,
to support the optimization of GCPV plants without increasing the costs
524–536. https://doi.org/10.1016/j.renene.2016.03.091.
of system design, making it beneficial in today’s highly competitive PV Bakhshi, R., Sadeh, J., Mosaddegh, H.R., 2014. Optimal economic designing of grid-
market. Besides the commercial application, the method can also be connected photovoltaic systems with multiple inverters using linear and nonlinear
module models based on Genetic Algorithm. Renew. Energy 72, 386–394. https://
useful for research purposes, e.g. in the assessment of the new tech-
doi.org/10.1016/j.renene.2014.07.035.
nologies like bifacial modules and floating PV systems or in the eva- Brandemuehl, M.J., Beckman, W.A., 1980. Transmission of diffuse radiation through CPC
luation of the effect of power forecast errors and energy storage in- and flat plate collector glazings. Sol. Energy 24, 511–513. https://doi.org/10.1016/
tegration. 0038-092X(80)90320-5.
Bryce, R., Losada, I., Kumler, A., Hodge, B., Roberts, B., Martinez-anido, C.B., 2018.
Consequences of neglecting the interannual variability of the solar resource : A case
7. Conclusion study of photovoltaic power among the Hawaiian Islands. Sol. Energy 167, 61–75.
https://doi.org/10.1016/j.solener.2018.03.085.
Cabrera-Tobar, A., Bullich-Massagué, E., Aragüés-Peñalba, M., Gomis-Bellmunt, O., 2016.
It has been proven that by application of the genetic algorithm ten Topologies for large scale photovoltaic power plants. Renew. Sustain. Energy Rev. 59,
important design parameters of grid-connected photovoltaic power

225
M.J. Mayer and G. Gróf Solar Energy 202 (2020) 210–226

309–319. https://doi.org/10.1016/j.rser.2015.12.362. Kornelakis, A., 2010. Multiobjective Particle Swarm Optimization for the optimal design
Cano, J., John, J.J., Tatapudi, S., Tamizhmani, G., 2014. Effect of tilt angle on soiling of of photovoltaic grid-connected systems. Sol. Energy 84, 2022–2033. https://doi.org/
photovoltaic modules. 2014 IEEE 40th Photovolt. Spec. Conf. PVSC 2014, 10.1016/j.solener.2010.10.001.
3174–3176. https://doi.org/10.1109/PVSC.2014.6925610. Kornelakis, A., Marinakis, Y., 2010. Contribution for optimal sizing of grid-connected PV-
Chen, S., Li, P., Brady, D., Lehman, B., 2013. Determining the optimum grid-connected systems using PSO. Renew. Energy 35, 1333–1341. https://doi.org/10.1016/j.
photovoltaic inverter size. Sol. Energy 87, 96–116. https://doi.org/10.1016/j. renene.2009.10.014.
solener.2012.09.012. Lindroos, J., Savin, H., 2016. Review of light-induced degradation in crystalline silicon
De Soto, W., Klein, S.A., Beckman, W.A., 2006. Improvement and validation of a model solar cells. Sol. Energy Mater. Sol. Cells 147, 115–126. https://doi.org/10.1016/j.
for photovoltaic array performance. Sol. Energy 80, 78–88. https://doi.org/10.1016/ solmat.2015.11.047.
j.solener.2005.06.010. Malamaki, K.N.D., Demoulias, C.S., 2014. Analytical calculation of the electrical energy
Dolara, A., Leva, S., Manzolini, G., 2015. Comparison of different physical models for PV losses on fixed-mounted PV plants. IEEE Trans. Sustain. Energy 5, 1080–1089.
power output prediction. Sol. Energy 119, 83–99. https://doi.org/10.1016/j.solener. https://doi.org/10.1109/TSTE.2014.2323694.
2015.06.017. Martins Deschamps, E., Rüther, R., 2019. Optimization of inverter loading ratio for grid
Duffie, J.A., Beckman, W.A., 2013. Sol. Eng. Therm. Processes. https://doi.org/10.1002/ connected photovoltaic systems. Sol. Energy 179, 106–118. https://doi.org/10.1016/
9781118671603.fmatter. j.solener.2018.12.051.
EC Joint Research Centre, 2017. Photovoltaic Geographical Information System 5 [WWW MathWorks, 2018. How the Genetic Algorithm Works [WWW Document]. Doc. R2018b.
Document]. URL http://re.jrc.ec.europa.eu/pvg_tools/en/tools.html. URL https://www.mathworks.com/help/gads/how-the-genetic-algorithm-works.
Erbs, D.G., Klein, S.A., Duffie, J.A., 1982. Estimation of the diffuse radiation fraction for html.
hourly, daily and monthly-average global radiation. Sol. Energy 28, 293–302. Mattei, M., Notton, G., Cristofari, C., Muselli, M., Poggi, P., 2006. Calculation of the
https://doi.org/10.1016/0038-092X(82)90302-4. polycrystalline PV module temperature using a simple method of energy balance.
Fu, R., Feldman, D., Margolis, R., 2018. U.S. Solar Photovoltaic System Cost Benchmark: Renew. Energy 31, 553–567. https://doi.org/10.1016/j.renene.2005.03.010.
Q1 2018, National Renewable Energy Laboratory. Golden. 10.7799/1325002. Notton, G., Lazarov, V., Stoyanov, L., 2010. Optimal sizing of a grid-connected PV system
Gallardo-Saavedra, S., Karlsson, B., 2018. Simulation, validation and analysis of shading for various PV module technologies and inclinations, inverter efficiency character-
effects on a PV system. Sol. Energy 170, 828–839. https://doi.org/10.1016/j.solener. istics and locations. Renew. Energy 35, 541–554. https://doi.org/10.1016/j.renene.
2018.06.035. 2009.07.013.
Gómez-Lorente, D., Triguero, I., Gil, C., Espín Estrella, A., 2012. Evolutionary algorithms Perez-Gallardo, J.R., Azzaro-Pantel, C., Astier, S., Domenech, S., Aguilar-Lasserre, A.,
for the design of grid-connected PV-systems. Expert Syst. Appl. 39, 8086–8094. 2014. Ecodesign of photovoltaic grid-connected systems. Renew. Energy 64, 82–97.
https://doi.org/10.1016/j.eswa.2012.01.159. https://doi.org/10.1016/j.renene.2013.10.027.
Gómez-Lorente, D., Triguero, I., Gil, C., Rabaza, O., 2014. Multi-objective evolutionary Rampinelli, G.A., Krenzinger, A., Chenlo Romero, F., 2014. Mathematical models for
algorithms for the design of grid-connected solar tracking systems. Int. J. Electr. efficiency of inverters used in grid connected photovoltaic systems. Renew. Sustain.
Power Energy Syst. 61, 371–379. https://doi.org/10.1016/j.ijepes.2014.03.064. Energy Rev. 34, 578–587. https://doi.org/10.1016/j.rser.2014.03.047.
Gómez, M., López, A., Jurado, F., 2010. Optimal placement and sizing from standpoint of Renewable Energy Policy Network for the 21st Century, 2018. Renewables 2018 Global
the investor of photovoltaics grid-connected systems using binary particle swarm Status Report. Paris. 978-3-9818911-3-3.
optimization. Appl. Energy 87, 1911–1918. https://doi.org/10.1016/j.apenergy. Schwingshackl, C., Petitta, M., Wagner, J.E., Belluardo, G., Moser, D., 2013. Wind effect
2009.12.021. on PV module temperature : Analysis of different techniques for an accurate esti-
Gracia, A.M., Huld, T., 2013. Performance comparison of different models for the esti- mation. Energy Procedia 40, 77–86. https://doi.org/10.1016/j.egypro.2013.08.010.
mation of global irradiance on inclined surfaces: Validation of the model im- Siddique, H.A.B., Xu, P., De Doncker, R.W., 2013. Parameter extraction algorithm for one-
plemented in PVGIS. European Commission, Joint Research Centre. https://doi.org/ diode model of PV panels based on datasheet values. 4th Int. Conf. Clean Electr.
10.2790/91554. Power Renew. Energy Resour. Impact, ICCEP 2013, 7–13. https://doi.org/10.1109/
Groniewsky, A., 2013. Exergoeconomic optimization of a thermal power plant using ICCEP.2013.6586957.
particle swarm optimization. Therm. Sci. 17, 509–524. https://doi.org/10.2298/ Sjerps-Koomen, E.A., Alsema, E.A., Turkenburg, W.C., 1996. A simple model for PV
TSCI120625213G. module reflection losses under field conditions. Sol. Energy 57, 421–432. https://doi.
Gulin, M., Vašak, M., Baotic, M., 2013. Estimation of the global solar irradiance on tilted org/10.1016/S0038-092X(96)00137-5.
surfaces. In: Proceedings of the 17th International Conference on Electrical Drives Smith, R.M., Jordan, D.C., Kurtz, S.R., 2012. Outdoor PV module degradation of current-
and Power Electronics, Dubrovnik, Croatia, 2–4 October. voltage parameters. In: Proceedings of World Renewable Energy Forum 2012.
Han, C., Lee, H., 2018. Investigation and modeling of long-term mismatch loss of pho- Denver, Colorado, p. 9.
tovoltaic array. Renew. Energy 121, 521–527. https://doi.org/10.1016/j.renene. Strobach, E., Faiman, D., Bader, S.J., Hile, S.J., 2013. Effective incidence angles of sky-
2018.01.065. diffuse and ground-reflected irradiance for various incidence angle modifier types.
Hay, J.E., 1993. Calculating solar radiation for inclined surfaces: Practical approaches. Sol. Energy 89, 81–88. https://doi.org/10.1016/j.solener.2012.10.021.
Renew. Energy 3, 373–380. https://doi.org/10.1016/0960-1481(93)90104-O. Sulaiman, S.I., Rahman, T.K.A., Musirin, I., Shaari, S., Sopian, K., 2012. An intelligent
Huld, T., Friesen, G., Skoczek, A., Kenny, R.P., Sample, T., Field, M., Dunlop, E.D., 2011. method for sizing optimization in grid-connected photovoltaic system. Sol. Energy
A power-rating model for crystalline silicon PV modules. Sol. Energy Mater. Sol. Cells 86, 2067–2082. https://doi.org/10.1016/j.solener.2012.04.009.
95, 3359–3369. https://doi.org/10.1016/j.solmat.2011.07.026. Vignola, F., Grover, C., Lemon, N., Mcmahan, A., 2012. Building a bankable solar ra-
Huld, T., Gracia Amillo, A.M., 2015. Estimating PV module performance over large diation dataset. Sol. Energy 86, 2218–2229. https://doi.org/10.1016/j.solener.2012.
geographical regions: The role of irradiance, air temperature, wind speed and solar 05.013.
spectrum. Energies 8, 5159–5181. https://doi.org/10.3390/en8065159. Vokony, I., Hartmann, B., Talamon, A., Papp, R.V., 2018. On selecting optimum tilt angle
Humada, A.M., Hojabri, M., Mekhilef, S., Hamada, H.M., 2016. Solar cell parameters for solar photovoltaic farms. Int. J. Renew. Energy Res. 8, 1926–1935.
extraction based on single and double-diode models: A review. Renew. Sustain. Winston, D.P., Kumar, B.P., Christabel, S.C., Chamkha, A.J., Sathyamurthy, R., 2018.
Energy Rev. 56, 494–509. https://doi.org/10.1016/j.rser.2015.11.051. Maximum power extraction in solar renewable power system - a bypass diode
IRENA, 2018. Renewable Power Generation Costs in 2017, International Renewable scanning approach. Comput. Electr. Eng. 70, 122–136. https://doi.org/10.1016/j.
Energy Agency. Abu Dhabi. 10.1007/SpringerReference_7300. compeleceng.2018.02.034.
Khatib, T., Mohamed, A., Sopian, K., 2013. A review of photovoltaic systems size opti- You, S., Lim, Y.J., Dai, Y., Wang, C.H., 2018. On the temporal modelling of solar pho-
mization techniques. Renew. Sustain. Energy Rev. 22, 454–465. https://doi.org/10. tovoltaic soiling: Energy and economic impacts in seven cities. Appl. Energy 228,
1016/j.rser.2013.02.023. 1136–1146. https://doi.org/10.1016/j.apenergy.2018.07.020.

226

You might also like