Energy Conversion and Management 174 (2018) 615–625

Contents lists available at ScienceDirect

Energy Conversion and Management

journal homepage: www.elsevier.com/locate/enconman

Effect of soiling in bifacial PV modules and cleaning schedule optimization T

⁎
Enric Grau Luque, Fernando Antonanzas-Torres, Rodrigo Escobar
Escuela de Ingeniería, Pontificia Universidad Católica de Chile, Vicuña Mackenna 4860, Santiago, Chile

A R T I C LE I N FO A B S T R A C T

Keywords: Bifacial photovoltaic (PV) modules and cells can transform solar radiation into electricity from both front and
Bifacial PV rear sides, unlike traditional solar technologies which can only generate power through the front face. This
Soiling ability has shown to increase electric output with various levels of increment, depending on parameters such as
Solar photovoltaic distance to the ground, distance between modules, and albedo. This power gain characteristic versatility has
Cleaning schedule
attracted the industry, gaining both commercial and research interest. Measuring soiling effects in bifacial
modules is an important milestone for the technology, since it is an important source of efficiency loss, thus
relevant to the industry when evaluating and designing bifacial systems. In this work soiling rates are measured
for bifacial minimodules and compared with traditional monofacial minimodules. The experiment was carried
out for a period of two months in Santiago, Chile, measuring short circuit current of the minimodules along with
the irradiance in the module plane and albedo. Also, a methodology is proposed to distinguish between soiling in
the front and rear sides of bifacial modules, with which a mixed integer lineal problem is designed to obtain
optimum cleaning frequency under different strategies and conditions for a period of three years, from 2014 to
2016. It is observed that soiling rate in the monofacial minimodule is 0.301%/day, meanwhile a rate of 0.236%/
day was measured for the bifacial module. Also, a rate of 0.0394%/day was calculated for the rear side of the
bifacial module, roughly 8.8 times smaller than the front side rate. Finally, several optimizations and simulations
where performed to see the effects of soiling rate, albedo, rain, cleaning costs and strategy in cleaning frequency
of both the front and rear sides of a bifacial PV system.

1. Introduction
Bifacial photovoltaic (PV) modules can transform solar radiation
into electricity from both front and rear sides, unlike traditional solar
modules which can only generate power through the front face. This
ability has shown to increase electric output with various levels of in-
crement, depending on parameters such as distance to the ground,
distance between modules, and albedo [1–3]. This power gain char-
acteristic has attracted the industry, gaining both commercial and re-
search interest.
Another reason for this growth in engagement with bifacial tech-
nology is the capacity to be installed vertically facing east-west. It has
been calculated that in this configuration is possible to increase power
output compared to a tilted configuration, depending on the geographic
location and albedo [4]. Also, when installed vertically facing east-west
the power peak is split in two and shifted to the morning and afternoon,
adjusting better to electric demands, lowering storage capacity [5].
Even though this technology has been studied since at least the early
60′s and scientifically described in the late 70′s [6], research and in-
dustry interest has concentrated in monofacial technology, dominating
the market with about 97% of the share in 2016, expecting bifacial
modules to reach 10% of market share by 2019 and over 35% in 2027
[7]. This lack of research in those first decades has resulted in a defi-
cient understanding of bifacial devices, making its entrance into the
market uncertain and slow. An example of this is the deficient char-
acterization of bifacial cells and modules in Standard Test Conditions
(STC). Though the method is accurate and reliable for traditional
technology, it has shown not to be adequate for bifacial devices, since
the method does not consider bifaciality. Some authors have tried to
counter this problem with innovative equipment [8,9], but further re-
search is needed to conceive to standard hardware and methodology
that outputs reliable and useful information. This characterization
problem also lead to a lack of simulation tools for bifacial modules,
meaning that it is not possible to generically simulate a bifacial solar
system, with some authors developing tools for only particular situa-
tions [10–12], being a problem for designers and engineers. Though, a
recent attempt tried to model generically bifacial systems [13], con-
sidering variables including module elevation, bifaciality and albedo
within a comprehensive. To validate their model they compare ex-
perimental and simulation results from other published articles

⁎
Corresponding author.
E-mail address: rescobar@ing.puc.cl (R. Escobar).

https://doi.org/10.1016/j.enconman.2018.08.065
Received 16 May 2018; Received in revised form 22 July 2018; Accepted 17 August 2018
Available online 24 August 2018
0196-8904/ © 2018 Elsevier Ltd. All rights reserved.
E.G. Luque et al.
including both tilted and vertical mounted bifacial systems. They obtain
remarkable low errors in their results with differences ranging from 0.1
to 4.4 percentage points in bifacial gain (i.e: difference between re-
ported bifacial gain in % and calculated bifacial gain in %). Never-
theless, further investigation and development is needed to lower this
error and develop a commercial available software that designers can
use and that also have better understanding and simulation of de-
gradation [14], thermal effects [15] and proper integration and simu-
lation of albedo [16].
Another matter with lack of research is soiling in bifacial modules.
Soiling is the accumulation of dust, dirt, snow, or any substance over
the photovoltaic module, decreasing its efficiency and therefore redu-
cing the electric output of the system [17–19]. It has also been docu-
mented that interest in soiling of solar technology, both thermal and
photovoltaic, has gained increasing interest over the past ten years,
pushed by manufacturing growth, decrease in module prices and in-
dustry interest [20]. As documented by Costa et al. [20], about 80
scientific articles related to soiling in solar systems were published in
2015, compared to about 10 articles in 2005.
The influence of soiling and it associated expenses have been stu-
died for different photovoltaic and thermal solar technologies in several
locations around the world, showing that the results vary for each
particular technology, location, climate, season and year [17,20,21].
Even though the problem can drastically change for each specific site,
the general solution is always cleaning, for what different methods can
be applied. This different cleaning methods can generally be classified
in four categories: natural cleaning, mechanical cleaning, self-cleaning
coatings, and electrostatic removal methods [22]. Of all the latter, the
second option is the most common solution in the industry, but self-
cleaning methods, particularly for glass, is a big scientific area, with
about 5000 published articles in 2017 and with a steady increase from
past years [23].
As there are associated costs to cleaning, a cleaning schedule opti-
mization should be prepared for each site and system, so the washing
costs do not surpass the economic benefit of it. A general and common
approach for this is to find an optimal cleaning interval considering
electricity sell price and cleaning costs. This is performed by Urrejola
et al. [24] using soiling data collected from a two-year experiment
performed in Santiago, Chile, analyzing different energy prices and
cleaning costs scenarios under a constant discount rate for a monofacial
polycrystalline residential system. The authors observe that for all cases
there is a critical cleaning period (CCP), representing the threshold day
from which no case should pass due to a rapid cleaning cost increase
beyond that day [24]. A similar approach is performed by Besson et al.
[25], where an optimization model is developed considering also the
effect of rain as natural free cleaning. The work shows how rain can
affect cleaning decisions of a solar system, being able to avoid me-
chanical cleaning when rain is considerable [25].
A different approach for defining an cleaning schedule is to use the
Levelized Cost of Energy (LCoE), as performed in [26]. In the article,
the authors use 21 months of data to evaluate the changes in LCoE due
to parameters such as soiling, degradation and temperature in thin film
and monocrystalline systems in the coastal zone of the Atacama Desert
in Chile. As a result, they conclude that at a rate of 12 cleanings a year,
the performance rate (PR) can improve 8% compared to three times a
year, improving the LCoE in 1.1 cent/kWp from 15.56 cent/kWp.
Another approach is to estimate optimal cleaning frequency de-
pending on the size of dust particles and accumulation density. The
latter is performed in [27], where a method is develop to estimate dust
deposition and cleaning frequency for different particle sizes and
module tilts in desert conditions. The simulated cases show that
cleaning time decreases with the increase of particle size and increases
with increasing angles. For the latter, when increasing the angle from 0°
to 80°, the optimal cleaning interval goes from 20 to 80 days when
particle size in 20 µm, and from 6,000 to 25,000 days when particle size
is 1 µm, thus site-specific dust characterization might be suitable to
616
Energy Conversion and Management 174 (2018) 615–625
properly design a cleaning schedule.
In terms of bifacial modules, it is known that the accumulation of
matter in a surface diminishes with the angle [28], so it is safe to say
that bifacial modules will suffer less soiling in the rear side when tilted
and in both sides when vertical facing east-west compared to tilted
monofacial modules. On the other hand, bifacial modules have two
sides to be cleaned, increasing the cost of a complete clean event,
compared to monofacial modules. Then it is of vital importance to
measure soiling rates for this technology and differentiate between
front and rear sides of the bifacial module and optimize cleaning
schedules considering all the above.
Only two references were found that included soiling for bifacial
modules. In the first work the authors compare the performance of
tilted mounted monofacial modules with vertical bifacial modules fa-
cing east-west in Saudi Arabia [29]. During the experimental period, a
dust storm occurred, by which the monofacial modules lost about 60%
in performance until they were cleaned, while the bifacial modules
didn't show any kind of loss. It must be considered that soiling during
dust storms is a very specific situation, thus the results cannot be
generalized and extrapolated to other types of dust, soiling and loca-
tions. Still, this occurrence is in accordance to the evidence and shows
an advantage of vertical bifacial modules.
Another work including soiling in bifacial modules found aimed to
evaluate the effects of the structural frame shade in energy yield of
bifacial modules, studying also soiling [30]. The authors installed tilted
monofacial and bifacial modules and vertical bifacial modules, mea-
suring short circuit current Isc . After 60 days of outdoor exposure the
modules were cleaned. To see soiling effects, they normalized the Isc
with global irradiance Gnet . It is calculated that the difference of this
ratio before and after the cleaning was basically 0% for vertical bifacial
modules, while the difference for tilted modules were about -12.5% for
the bifacial modules and -17.25% for the monofacial module. This work
then is also in accordance with evidence and consistent with results
observed in Hajjar et al. [29].
This work is focused on soiling analysis for bifacial modules in an
attempt to differentiate with traditional PV technology in the matter.
For this, a methodology to obtain soiling rates for the rear face from
experimental data is proposed, using the information to optimize the
cleaning schedule for both front and rear faces of a bifacial PV system,
analyzing different scenarios of albedo, cleaning costs, rain and
cleaning strategy. The paper is structured as follow: chapter 2 collates
the experimental setup and procedure, chapter 3 the optimization
model, chapter 4 the results and discussion and eventually, chapter 5
presents the conclusions.
2. Experimental setup and procedure
2.1. Installation and site description
This study considered two polycrystalline monofacial minimodules
and two bifacial monocrystalline minimodules with four cells in series
each, as shown in Fig. 1. For each minimodule, short circuit current Isc
was measured through a shunt resistor. Global irradiance in the plane of
the array GPOA and albedo A were measured with pyranometers. The
albedo was measured at 180° in the same axis as the GPOA pyranometer
(inverse plane). Data was collected in a time resolution of one minute.
An image of the setup is presented in Fig. 1. For each technology, one
module was assigned to be cleaned once a week, while the other was
left to soil through the whole experimental period. The modules where
tilted at 35° and 10° azimuth (North = 0°).
The study was performed on the solar laboratory at Pontificia
Universidad Católica de Chile (PUC), in the roof of a three-story
building inside the campus. Measurements started on September 12th,
2017 and concluded on November 21st. The university campus is lo-
cated in the city of Santiago, the capital of Chile. According to the last
census of 2017 [31], the city has a population of 5.2 million people,
E.G. Luque et al.
Fig. 1. Experimental equipment used: on the left a general view of the structure, four modules and pyranometers and at the right a close-up view of the POA and
Albedo pyranometers setup. The experiment was performed in the roof of a three-story building and later moved to a new facility after the experiment was finished.
The image corresponds to the new laboratory at the roof of a seven-story building for display purposes only.
about 30% of the entire population, indicating then to be the city with
biggest demand for residential PV systems, installed at houses and
buildings rooftops.
Geographically speaking, the city of Santiago is located between
two mountain ranges: Cordillera de los Andes and Cordillera de la
Costa. This condition obstructs air flow and air renovation, favoring the
accumulation of different pollutants such as CO, SO2 and particulate
matter (PM) PM10 and PM2.5, exceeding constantly air quality standards
[32]. It has been calculated that the mixing layer height (MLH), the
limit where pollutants are mixed with air by turbulence and diffusion,
can range around 1,800 m at any given day [33]. Even though no
chemical analysis of the accumulated soil on the modules was per-
formed, soiling properties can be related to the pollution levels due to
this thick mixing layer of pollutants and its horizontal and vertical
diffusion properties, hence similar effects could be observed through
the city at different neighborhoods and altitudes on residential con-
structions. This pollution layer can be easily appreciated in Fig. 1
during an average spring day
The climate in Santiago is considered Mediterranean, with a long
dry summer station and rainy winter. During the summer, temperatures
can reach over 33 °C and drop below 0° in winter. Rainfall will typically
concentrate during the months of May through June, and some pre-
cipitation may be observed during the summer period. Windspeed does
not change significantly through the year, though in tends to be lower
in winter with an east predominant direction. Daily average values for a
Typical Meteorological Year (TMY) for GHI, DNI, temperature, rain,
relative humidity, and wind speed are plotted in Fig. 2, indicating
maximum and minimum values for each variable. Data was extracted
from the Solar Explorer of the Ministry of Energy of Chile1.
2.2. Data processing
In order to use more representative data, a set of filters were applied
to rule out measurements. In particular, the criteria is presented in
Table 1, as performed in Besson et al. [25]. With this, days 39 and 51
were ruled completely due to heavy cloud density.
In Fig. 3, the ratio RIG = Isc, meadured / GPOA is plotted for each mini-
module, being Isc, measured the measured short circuit current, as per-
formed in [34]. In Fig. 4 the relative difference dr for each technology is
shown, defined as
RIG, clean−RIG, soiled
dr = × 100
RIG, clean (1)
Where RIG, soiled and RIG, clean stand for the ratio between the short circuit
1
Website: http://www.minenergia.cl/exploradorsolar/
617
Energy Conversion and Management 174 (2018) 615–625
current of the soiled module and the clean module, respectively.
For each technology, dr is shifted so it is zero for the first day (i.e.:
dr,0 = 0 ), as shown in Fig. 4. Comparison between GPOA and albedo A
and respective averages are presented in Fig. 5.
2.3. Soiling rate
Soiling rate (SR) is calculated by subtracting the relative difference
in day a (dr , a ) and day b (dr , b ), dividing by the number of days b−a . Due
to rain events, the experiment period is divided in three sections, thus
calculating three soiling rates. Between days 19 and 24, heavy rain
events cleaned completely the front of all the modules, and a small rain
event in day 51 cleaned partially the front of the modules. With the
latter, considered periods and respective soiling rates for the bifacial
and monofacial minimodules are presented in Table 2.
To fully characterize soiling in bifacial modules, it should be dis-
tinguished between the front and rear faces. For this, it is possible to
express the ideal short circuit current of the module Isc as
Isc = If + Ir (2)
where If and Ir are the ideal short circuit currents contributed by the
front and rear sides without soiling, respectively.
Also, bifacial modules might have different efficiencies in each side,
being the front more efficient than the rear side, then it is possible to
write the rear current as
Ir = If × B (3)
where B is the bifaciality of the device, defined as
ηr
B=
ηf (4)
being ηf and ηr the efficiency of the front a rear side, respectively.
Also, in bifacial devices the rear side receives albedo A , defined as a
fraction of the irradiance, so equation (2) can be rewritten as
Isc = If + A × B × If (5)
Isc
If =
1+A×B (6)
Then, when soiling is present, it is described as a rate in a time scale
(day, month, year) multiplied by the time passed since last cleaning,
thus:
Isc, measured = If × (1−SRf × Nf ) + Ir ·(1−SRr ·Nr ) (7)
Isc
Isc, measured = × ((1−SRf × Nf ) + A × B × (1−SRr × Nr ))
1+A×B
(8)
E.G. Luque et al. Energy Conversion and Management 174 (2018) 615–625

Fig. 2. climate conditions for GHI, DNI, temperature, relative humidity, rain, and wind speed in daily averages for a TMY in Santiago, Chile. Maximum and minimum
values for each variable are labeled.

Table 1 rewritten as
Parameter acceptance criteria for collected data during the experiment period.
RIG
Parameter Acceptance criteria Unit RIG, measured = × ((1−SRf × Nf ) + A × B × (1−SRr × Nr ))
1 + A·B (9)
GPOA X > 300 W/m2
With this, values for SRr are resumed in Table 3. Soiling rates for the
Solar azimuth −80 < X < 80 °
Solar elevation X > 25 ° bifacial moduleSRb , monofacial module SRm and for the rear side of the
bifacial module SRr are plotted in Fig. 6.

Fig. 3. Comparison between RIG, clean and RIG, soiled for both monofacial and bifacial minimodules. Dashed lines show the linear regression performed for each set of
data.

where SRf and SRr are the soiling rates for the front and rear sides of the
bifacial module, respectively, Nf and Nr are the time without cleaning
for each side, and Isc, measured is the measured current.
To estimate the soiling rate SRr for the rear side of the bifacial
minimodule, experimental data is introduced in equation (8) divided by
GPOA so it is in function of RIG instead of the short circuit current. Soiling
rate of the front SRf is assumed to be equal to the soiling rate of the
monofacial module (SRf = SRm ) and B = 1. With the latter, (8) is
2.4. Validation
In order to validate Eq. (9), simulations were performed for the first
49 days introducing experimental data acquired during the experiment.
This was performed because the equation supposes known values for Nf
and Nr , which is not possible after the rain events in day 50, where the
rain cleaned significantly the modules but not completely, thus the
equivalent days of soiling it is not known.

618
E.G. Luque et al. Energy Conversion and Management 174 (2018) 615–625

Fig. 4. Relative difference dr for both bifacial and monofacial minimodules. Dashed lines show the linear regression performed for each set of data.

Fig. 5. Comparison of GPOA , albedo A and respective averages.

Table 2 To validate the lineal regression, the determination coefficient (R2)

Soiling rates for bifacial and monofacial minimodules. was used, as seen in Table 4, including also the correlation coefficient
Day a Day b dr , a (%) dr , b (%) dr , a−dr , a Days Soiling SRb (%/day) and root mean square error (RMSE) as a reference. It was obtained a R2
of 0.84 for both clean and soiled bifacial module RIG (Fig. 3). Also, a
Bifacial module 0.89 was obtained for the relative difference, as seen in Fig. 4. On the
1 15 0 3.74 3.74 14 0.249 other hand, for the clean monofacial a 0.63 value was calculated.
25 46 0.310 4.81 4.50 22 0.204
Meanwhile, the R2 corresponding to the relative difference for the
52 64 2.13 5.21 3.08 13 0.238
Average: 0.236 monofacial modules achieved a 0.94 value.
Monofacial module
1 15 0 3.90 3.90 14 0.279 3. Optimization model
24 46 −0.07 4.98 5.05 22 0.223
51 64 1.49 6.71 5.22 13 0.402 The optimization model is based in the approach by Besson et al.
Average: 0.301
[25]. In this publication, they design an optimization problem for
cleaning frequency of a Mono c-Si 1590 Wp system. In particular, what
they proposed is the following mixed integer lineal problem:
Table 3
Estimated soiling rates for the rear face of the bifacial module. T
max ∑ sst × CPt × EP−x t CC
Bifacial module, rear side (10)
t=1

Day a Day b Nf Nr SRr (%/day) Subject to:

1 15 14 14 0.0429 sst + 1 ≤ sst −SRt × (1−x t ) + x t + Rt (11)

24 46 22 46 0.0357
51 64 13 64 0.0396 T
Average: 0.0394 MinC ≤ ∑ xt ≤ MaxC
t=1 (12)

0 ≤ sst ≤ 1 (13)
To estimate the ideal RIG a lineal regression was made considering
only clean days of the bifacial and monofacial modules. Results of the where the maximized function is the balance between electricity cost
simulations are presented in Fig. 3 and Fig. 4 as dashed lines. and cleaning costs, sst is the soiling state, being 100 completely cleaned

619
E.G. Luque et al. Energy Conversion and Management 174 (2018) 615–625

Fig. 6. Soiling rates during the experimental period for the monofacial and bifacial modules (left) and for the rear side of the bifacial module (right) with respective
averages.

Table 4 binary cleaning decision parameters for each side of the bifacial system,
Correlation coefficient, R2 coefficient and RMSE for each simulation, between similar to what Besson et al. [25] did for a monofacial PV system.
days 1 and 50.
Sim Corr. Coef. R2 Coef. RMSE 3.1. Daily yield

Bifi clean 0.91 0.84 0.00013 The daily yield CPft was obtained from the simulation of a 1 kWp
Bifi soiled 0.91 0.84 0.00014
monofacial system in Santiago, Chile, using the software SAM [35] and
Bifi diff 0.94 0.89 0.58
Mono clean 0.79 0.63 0.00012 solar resource data for 2014–2016 measured in the same laboratory as
Mono soiled 0.86 0.75 0.00013 the experiment. From that, CPrt is in function of CPft , this because no
Mono diff 0.97 0.94 0.63 reliable enough tool for simulating bifacial systems has been developed,
being this a simple deterministic approach for bifacial gain in power
yield.
and 0 completely soiled, x t the cleaning decision binary variable for
each day t , being 1 for cleaning and 0 for not cleaning, CPt the daily 3.2. Cleaning cost
ideal energy yield not affected by soiling in kWh/day, EP a constant
price of 160 USD/kWh at which the energy is sold, CC the cleaning cost In Besson et al. [25] the CC was 1 and 4 USD/kW representing a
in USD/kWp/clean varying two values of 1 and 4, SRt the soiling rate in low-cost and high-cost cleaning scenario, respectively. For this work the
%/day, Rt a binary variable being 1 if rain is over 1 mm (considered a same procedure was performed, with the additional consideration that
free perfect cleaning) and 0 for days without rain, MinC and MaxC the the CC is equal for both sides. As a result it cannot be concluded either a
minimum and maximum allowed cleanings for the study horizon, t in higher or lower value for rear cleaning, since much lower amounts of
days, and T as the study horizon in days, being 365 for the year 2014 dirt are accumulated, thus it might take less effort to clean the rear. On
[25]. the other hand, the difficulty of cleaning the rear side of the modules
The latter does not consider bifaciality, so it is necessary to consider might increase costs, but no evidence or information on this issue or the
it in order to obtain an optimal frequency for bifacial systems. In par- latter is described in the literature. Also, it is considered that rain does
ticular, it is a requirement to distinguish between the front and rear face not affect the soiling state in the rear side, since it is covered by the
of the bifacial module, thus the problem is modified as follows: front, and bouncing rain drops from the frame, condensation and wind
T [36] in the rear side are not accounted to have cleaning or any addi-
max ∑ (ssft × CPft + ssrt × B × A × CPft ) × EP−CC (xft + xrt ) tional effects.
t=1 (14)
Subject to: 3.3. Albedo
ssft + 1 ≤ ssft −SRft × (1−xft ) + xft + Rt (15)
During the experiment an albedo value of 0.12 was measured, being
ssrt + 1 ≤ ssrt −SRrt × (1−xrt ) + xrt (16) the unaltered albedo at the laboratory. Higher albedo magnitudes are
fairly common on the literature [37–40], so it makes sense to increase
T
the albedo to observe the effects on the model. For this, an albedo of
MinC ≤ ∑ xft ≤ MaxC
0.24 was also analyzed, being this the double of the measured albedo in
t=1 (17)
site. Though, 0.24 is easily obtainable with different materials available
T
in nature and the market, as shown in [41] with materials such as grey
MinC ≤ ∑ xrt ≤ MaxC and white shingles, concrete, sandstone, sand and other materials sui-
t=1 (18)
table for residential and industrial use.
0 ≤ ssft ≤ 1 (19)
3.4. Electricity price
0 ≤ ssrt ≤ 1 (20)
where B is the bifaciality of the device, A the albedo, ssft and ssrt are the In Besson et al. [25], a constant EP of 160 USD/kWh was considered
soiling states for the front a rear sides, CPft the ideal energy yield for the one-year optimizations. In this work, as more years are con-
without soiling, xft and xrt the cleaning decision binary variables for sidered, using an average price comes as an important assumption since
each day t and each side, SRft and SRrt the soiling rates in in %/day, they change every month. In Chile, regulated tariffs are divided in low-
MinC , MaxC, MinC ', and MaxC' the minimum and maximum allowed tension (BT) and high-tension (AT), having both options different
cleanings for the study horizon. In short, this problem optimizes the billing schemes and prices for electricity, power, and other charges. A
balance between electric prices and cleaning costs, by optimizing the common tariff for residential bills is BT-1, which charges only a fixed

620
E.G. Luque et al. Energy Conversion and Management 174 (2018) 615–625

Table 5 frequency for once a month (35 cleanings), every two months (17
Specifications for each optimization and simulation and number of cleanings cleanings), every three months (11 cleanings) and every four months (8
events for the front (FC) and rear (RC) and Balance (Bal). cleanings) for the best scenario in terms of cleaning costs and albedo of
With rain Without rain CC = 1 USD/kW and A = 0.12 . In these cases, when rain was con-
sidered, no cleaning was performed for months in which rain was over
Tech CC ($) A N° FC RC Bal. ($) N° FC RC Bal. ($) 1 mm, considered as free cleaning. The results are summarized in
Table 5.
Separate cleaning
Bifi 1 0.12 1.a 15 3 805 5.a 34 4 782 xrt = xft (21)
Bifi 4 0.12 2.a 2 1 777 6.a 16 1 709
Bifi 1 0.24 3.a 15 5 893 7.a 33 5 869
xrt = 0 (22)
Bifi 4 0.24 4.a 2 2 863 8.a 15 2 794

Simultaneous cleaning
Bifi 1 0.12 1.b 10 10 799 5.b 23 23 760 4. Results and discussion
Bifi 4 0.12 2.b 2 2 772 6.b 11 11 668
Bifi 1 0.24 3.b 10 10 889 7.b 24 24 852
Bifi 4 0.24 4.b 3 3 858 8.b 12 12 759 In this section, results are discussed for the number of optimizations
and simulations performed. To ease the latter, each optimization and
No rear cleaning
Bifi 1 0.12 1.c 15 0 794 5.c 33 0 771
simulation has been assigned with an identifier consisting in a number
Bifi 4 0.12 2.c 2 0 772 6.c 16 0 704 and a lower-case letter separated by a dot. When indicating an X instead
Bifi 1 0.24 3.c 15 0 867 7.c 33 0 848 of the number or letter, it references to all the cases with the explicit
Bifi 4 0.24 4.c 2 0 844 8.c 15 0 777 character. For example, 7.x indicates all cases that contain “7″ (7.a, 7.b
Simulations and so on), and X.a all cases that contain “a” (1.a, 2.a and so on). Every
Bifi 1 0.24 1.d 15 15 881 9.d 35 35 843 case is referenced to with this nomenclature, which is indicated in
Bifi 1 0.24 2.d 15 0 862 10.d 35 0 841
Table 5 as N°.
Bifi 1 0.24 3.d 9 9 881 11.d 17 17 843
Bifi 1 0.24 4.d 9 0 856 12.d 7 0 824
Bifi 1 0.24 5.d 5 5 871 13.d 11 11 817 4.1. Optimization cases
Bifi 1 0.24 6.d 5 0 849 14.d 11 0 794
Bifi 1 0.24 7.d 4 4 868 15.d 8 8 788
Bifi 1 0.24 8.d 4 0 843 16.d 8 0 762
As expected, the highest balance was obtained when cleaning cost
CC was lower and albedo A is higher, what was seen for each strategy
Monofacial
in cases 3.x (considering rain) and 7.x (without rain). On the contrary,
Mono 1 – 1.e 15 – 721 3.e 34 – 698
Mono 4 – 2.e 2 – 699 4.e 16 – 630 the lowest balance was when CC was higher and albedo lower, found in
cases 2.x and 6.x. In particular, the highest balance with rain was ob-
tained in case 3.a with $893 and the lowest in case 2.b and 2.c with
price (service) and electricity. Taking into account the limitations of $772, a 13.6% or $121 lower. This last situation stands out since both
this model, as it is more suitable for a small residential system, the BT-1 2.b and 2.c have A = 0.12 and CC = 4 , but in 2.b a total of four cleaning
electricity prices are considered for each month, through 2014 to 2016, events are performed (two in each side), meanwhile in 2.c only one in
for the location of the simulated system at PUC. the front, thus obtaining equal results with less activity. This is illu-
strated in Fig. 8A, in addition to simulation 2.a. It can be observed that
front cleaning events vary slightly for 2.b, meanwhile 2.a and 2.c pre-
3.5. Cleaning strategies sent identical cleaning schedule for the front side and 2.a obtaining a
modest $5 higher balance of $ 777 (see Fig. 9).
In addition to the latter, different cleaning strategies were evaluated When increasing the albedo from 0.12 to 0.24 and maintaining CC
for the bifacial system, being this: separate cleaning (modified model as and the strategy, the largest balance gain was observed between 6.b and
shown), simultaneous cleaning (by adding Eq. (21) as restriction), no 8.b of about 12.0% or $91 with 11 and 12 cleanings in each side, re-
cleaning on the rear side (adding Eq. (22) as restriction) and a steady spectively. The lowest increase was observed between 1.c and 3.c of
frequency for both sides and only the front. A total of 44 analyses were 8.42% or $73 with no change in FC (RC = 0). Overall, the average
performed for the years 2014–2016. Out of these 44 cases, 28 are op- increase in balance by changing the albedo from 0.12 to 0.24 was
timizations using the above model and 16 are simulations at steady 10.9%.

Fig. 7. Sensibility analysis showing balance with low A values and high soiling rates for the rear face of the bifacial module. Monofacial lines at CC = 1 and CC = 4
are constant.

621
E.G. Luque et al.
Fig. 8. Results of optimization model for a sample of optimizations in terms of Soiling State (SS) and Balance. Use Table 5 to reference exact parameters for each
simulation number.
When increasing the cleaning cost CC from 1 to 4 and maintaining
albedo and strategy, the largest balance difference was observed be-
tween simulations 5.b and 6.b with 12.1% or $92 changing FC and RC
from 23 to 11, while the lowest difference was found between 3.c and
4.c with 2.64% or $23 changing FC from 15 to 2 (RC = 0). In com-
parison, for the monofacial system there was a 3.1% or $22 difference
between 1.e and 2.e changing FC from 15 to 2, and a 9.7% or $68
difference between 3.e and 4.e changing FC from 34 to 16. For the
bifacial system, the average decrease in balance when changing CC
from 1 to 4 was 3.2% or $27 with rain, and 9.3% or $75 without rain.
When comparing bifacial with monofacial technology it can be
appreciated that there are small differences. In fact, when considering
rain, the difference in front cleanings was 1 between 1.a and 1.e and no
622
Energy Conversion and Management 174 (2018) 615–625
difference between 2.a and 2.e, being the front cleaning frequency
virtually the same. In this case, the biggest difference was in the balance
values with a 10.6% or $85 difference between 1.a and 1.e and 10.0%
or $78 difference between 2.a and 2.e. In the same way, without con-
sidering rain, the difference lays in the balance, being 10.7% or $84
higher in 5.a respect to 3.e and both with 34 front cleanings. Also, 6.a
was 11.1% or $79 higher respect to 4.e, but this with the bifacial si-
mulation adds one cleaning over the 16 in 10.b, thus showing a small
difference in the three-year period considered.
In general, best balance values are obtained when cleaning was
optimized separately, followed by simultaneous cleaning being in
average 0.57% lower with rain and 3.2% without rain. With no rear
cleaning the balance was reduced in average 1.5% with rain and 1.5%
E.G. Luque et al.
Fig. 9. Comparison for a sample of simulations and optimization in terms of Soiling State (SS) and Balance. Use Table 5 to reference exact parameters for each
simulation number.
without rain. The latter is illustrated in Fig. 8C, presenting simulations
with same parameters without rain. In the figure, the balance lines
show little difference while having significantly different cleaning ac-
tivities in both front and rear sides. The difference was even lower when
rain was present, this because the rain events allow for more similar
cleaning schedule on the front face.
Overall, it is clear that albedo has an important effect on the balance
value, having a greater effect than the cleaning cost. On the other hand,
cleaning costs have a greater effect on frequency, especially when rain
is considered. With all this in mind it is evident that, though cleaning
the front and rear sides according to the separate cleaning strategy has
higher balance values, the difference with simultaneous and no rear
cleaning strategies of 0.63% and 1.5% (with rain) are low enough that
by adding other variables, such as logistics, might turn this results
around. This is because it is easier to schedule a simultaneous cleaning
event or just ignore rear cleaning than planning and performing a se-
parate cleaning schedule.
4.2. Comparing optimization with simulation cases
After the optimizations, simulations generated for once a month,
every two, three and four months cleaning were performed for the best
scenario in terms of cleaning costs and albedo of CC = 1 USD/kW and
A = 0.24 . Cleaning was performed in last day of the corresponding
months if daily rain events were under 1 mm during that month. When
analyzing the simulation cases, it can be seen that for all cases balances
were higher when both rear and front cleanings are performed, except
for 9.d and 10.d. Also, it shows that all optimized scenarios with same
623
Energy Conversion and Management 174 (2018) 615–625
CC and A balances were higher. It is interesting to notice simulation 2.d
with 15 front cleaning events and 0 rear cleanings has the same
cleaning event number as 3.c, but this latter having a balance of $867
and the first $862, 0.58% smaller, as illustrated in Fig. 8A. Just as
observed in the optimizations, rain has an important role, reducing
necessary cleanings by 63% when optimizing and 53% for the simula-
tions, in average. In terms of balance, 1.d and 3.d have virtually same
value of $881, though the last one with 67% less cleanings. Similarly,
3.d is comparable with 3.b, that with two less cleaning events it ob-
tained a 0.90% smaller balance, and 1.5% smaller than 3.a with one less
cleaning event (Fig. 9).
4.3. Sensibility analysis
When varying the soiling rate SSr from 0.03 to 0.1 it was seen that,
under same cleaning cost CC the balance was always greater for a bi-
facial system (Fig. 7). In the worst case simulated with CC = 4,
A = 0.03 and SSr = 0.104 the balance was $715, meanwhile the bal-
ance for a monofacial system under same conditions is $699, 2.2%
lower. This is due to the particular way in which the problem was
modeled, since bifacility is considered as a gain over the monofacial
yield. Thus, it always generates more power even if no rear cleanings
were performed.
5. Conclusions
This work analyses the outdoor performance of monocrystalline
bifacial mini-modules and polycrystalline monofacial minimodules in
E.G. Luque et al.
terms of soiling by measuring the short circuit current for a period of
two months in Santiago, Chile. The main purpose of the investigation
was to characterize soiling for bifacial devices and differentiate from
traditional technologies. For the latter, a set of equations and metho-
dology are proposed to distinguish soiling on the front and rear sides of
the bifacial module. With the equations, average soiling rate of
0.301%/day is calculated for the monofacial minimodule, meanwhile a
rate of 0.236%/day is estimated for the bifacial module, being then
21.6% lower. For the rear side of the bifacial module a rate of 0.394%/
day was estimated, roughly 7.6 times smaller compared to the mono-
facial soiling rate. Then, the equations were simulated using linear re-
gression and soiling data obtained from the two-month period experi-
ment, demonstrating agreement between the simulated and
experimental data. Finally, a cleaning optimization model from Besson
et al. [25] was modified to consider bifaciality, and several simulations
were performed allowing to quantify the effects of albedo, cleaning
costs, rain, cleaning frequency and strategy in a bifacial system. It is
observed that when optimizing for each side separately, the balance is
greater by less than 3% if optimized for simultaneous cleaning or
without cleaning the rear side at all in a three-year period. This rela-
tively low difference suggests that by including other variables into the
equation could turn results around. Thus, ignoring cleaning on the rear
could be already a viable and acceptable strategy under presented
conditions and results, leading to be more profitable in particular si-
tuations. When comparing the optimizations with the simulations of
simple schedules of 12, 6, 4 and 3 times a year, it is clear that opti-
mizing is an important exercise, having always better balances in the
optimized cases. In general, using the optimization tool decreases the
amount of cleaning events and increases balance for the bifacial system,
and optimizing rear cleaning separately is the optimum choice of the
presented strategies, thus is concluded that soiling in the rear side is an
issue that must be considered and studied further.
6. Future work
As mentioned, the model is a simple deterministic way to approach
cleaning decisions in a bifacial PV system. In particular, the model does
not consider the cleaning time of the modules or system, thus it is
limited to analyze residential systems that are small enough to be
cleaned in less than one day, the smallest time scale at which this work
is done. Also, it is important to notice that the model makes an optimum
schedule for an extended period of three years using meteorological
data, thus the results should be used as reference only for decision
making in the short-term and to compare with steady-interval cleaning
schedules. With this in mind, several points arise to be addressed in the
future. First, full-size modules should be studied, as different effects can
occur due to a larger glass surface area, cell-strings with diodes, and
accumulation of dust in the module frame. As described in this work,
soiling rate for the rear side is calculated from the module soiling rate,
so the rear soiling rate should be measured directly with a different
methodology, preferably measuring the I-V curve or the Performance
Ratio. Also, as mentioned in the Introduction, vertical bifacial modules
are being studied, thus soiling in that configuration should be assessed
in detail. Measuring soiling rates in different weather conditions is also
of high interest, especially in desertic conditions as in the Atacama
Desert in Chile. Finally, soiling in bifacial modules in a tracking rack
configuration is of high interest due to the increase in power output
while tracking the sun in 1-axis systems.
Acknowledgments
The authors acknowledge and thanks the Energy Research Centre of
the Netherlands (ECN) for providing the monofacial and bifacial
minimodules, items with which this work wouldn't have been possible.
Also, the authors acknowledge to the Engineering School of Pontificia
Universidad Católica de Chile for its postdoctoral support which
624
Energy Conversion and Management 174 (2018) 615–625
enabled this research.
References
[1] Yusufoglu UA, Pletzer TM, Koduvelikulathu LJ, Comparotto C, Kopecek R, Kurz H.
Analysis of the annual performance of bifacial modules and optimization methods.
IEEE J Photovolt 2015;5(1):320–8.
[2] Kreinin L, Bordin N, Karsenty A, Drori A, Grobgeld D, Eisenberg N. PV Module
Power Gain due to Bifacial Design. Premliminary Experimental and Data; 2010.
[3] Appelbaum J. Bifacial photovoltaic panels field. Renew Energy 2016;85:338–43.
[4] Guo S, Walsh TM, Peters M. Vertically mounted bifacial photovoltaic modules: A
global analysis. Energy 2013;61:447–54.
[5] Obara S, Konno D, Utsugi Y, Morel J. Analysis of output power and capacity re-
duction in electrical storage facilities by peak shift control of PV system with bi-
facial modules. Appl Energy 2014;128:35–48.
[6] Cuevas A, Luque A, Ruíz JM. Bifacial Transcells for Luminiscent Solar
Concentration. In: Electron Devices Meeting, 1979, p. 314–17.
[7] ITRPV. International Technology Roadmap for Photovoltaic Seventh Edition, March
2017. Frankfurt, Germany, 2017.
[8] Ohtsuka H, Sakamoto M, Koyama M, Tsutsui K, Uematsu T, Yazawa Y.
Characteristics of bifacial solar cells under bifacial illumination with various in-
tensity levels. Prog Photovolt Res Appl 2001.
[9] Ezquer M, Petrina I, Cuadra JM, Lagunas AR. Design of a special set-up for the I-V
characterization of bifacial photovoltaic solar cells. 23rd Eur Photovolt Sol Energy
Conf Exhib 2008.
[10] Lo CK, Lim YS, Rahman FA. New integrated simulation tool for the optimum design
of bifacial solar panel with re fl ectors on a specific site. Renew. Energy 2015;81.
[11] Janssen GJM, Van Aken BB, Carr AJ, Mewe AA. Outdoor performance of bifacial
modules by measurements and modelling. Energy Procedia 2015;77:364–73.
[12] Castillo-Aguilella JE, Hauser PS. Multi-variable bifacial photovoltaic module test
results and best-fit annual bifacial energy yield model. IEEE Access
2016;4:498–506.
[13] Sun X, Khan MR, Deline C, Alam MA. Optimization and performance of bifacial
solar modules: A global perspective. Appl Energy 2018;212(September
2017):1601–10.
[14] Huang C, Wang L. Simulation study on the degradation process of photovoltaic
modules. Energy Convers Manage 2018;165(March):236–43.
[15] Abd-Elhady MS, Serag Z, Kandil HA. An innovative solution to the overheating
problem of PV panels. Energy Convers Manage 2018;157(December 2017):452–9.
[16] Calinoiu D, Stefu N, Boata R, Blaga R, Pop N, Paulescu E, Sabadus A, Paulescu M.
Parametric modeling: A simple and versatile route to solar irradiance. Energy
Convers. Manage 2018;164(December 2017):175–87.
[17] Maghami MR, Hizam H, Gomes C, Radzi MA, Rezadad MI, Hajighorbani S. Power
loss due to soiling on solar panel: A review. Renew Sustain Energy Rev
2016;59:1307–16.
[18] Xu R, Ni K, Hu Y, Si J, Wen H, Yu D. Analysis of the optimum tilt angle for a soiled
PV panel. Energy Convers Manage 2017;148:100–9.
[19] Pulipaka S, Kumar R. Analysis of irradiance losses on a soiled photovoltaic panel
using contours. Energy Convers Manage 2016;115:327–36.
[20] Costa SCS, Diniz ASAC, Kazmerski LL. Dust and soiling issues and impacts relating
to solar energy systems: Literature review update for 2012–2015. Renew Sustain
Energy Rev 2016;63:33–61.
[21] Sarver T, Al-Qaraghuli A, Kazmerski LL. A comprehensive review of the impact of
dust on the use of solar energy: History, investigations, results, literature, and mi-
tigation approaches. Renew Sustain Energy Rev 2013;22:698–733.
[22] He G, Zhou C, Li Z. Review of self-cleaning method for solar cell array. Procedia Eng
2011;16:640–5.
[23] Syafiq A, Pandey AK, Adzman NN, Rahim NA. Advances in approaches and methods
for self-cleaning of solar photovoltaic panels. Sol Energy 2018;162(November
2017):597–619.
[24] Urrejola E, Antonanzas J, Ayala P, Salgado M, Ramírez-Sagner G, Cortés C, et al.
Effect of soiling and sunlight exposure on the performance ratio of photovoltaic
technologies in Santiago, Chile. Energy Convers Manage 2016;114:338–47.
[25] Besson P, Munoz C, Ramirez-Sagner G, Salgado M, Escobar R, Platzer W. Long-term
soiling analysis for three photovoltaic technologies in Santiago Region. IEEE J
Photovolt 2017;7(6):1755–60.
[26] Fuentealba E, Ferrada P, Araya F, Marzo A, Parrado C, Portillo C. Photovoltaic
performance and LCoE comparison at the coastal zone of the Atacama Desert, Chile.
Energy Convers Manage 2015;95:181–6.
[27] Jiang Y, Lu L, Lu H. A novel model to estimate the cleaning frequency for dirty solar
photovoltaic (PV) modules in desert environment. Sol Energy 2016;140.
[28] Elminir HK, Ghitas AE, Hamid RH, El-Hussainy F, Beheary MM, Abdel-Moneim KM.
Effect of dust on the transparent cover of solar collectors. Energy Convers Manage
2006;47(18–19):3192–203.
[29] Hajjar HK, Dubaikel FA, Ballard IM. Bifacial photovoltaic technology for the oil and
gas industry. 2015 Saudi Arab. Smart Grid, SASG 2015, Apr. 2016.
[30] Rabanal-arabach J, Mrcarica M, Schneider A. The need of frameless mounting
structures for vertical mounting of bifacial PV modules. 32nd EU PVSEC 2016.
[31] INE, Síntesis de Resultados Censo 2017; 2018.
[32] Jorquera H. Air quality at Santiago , Chile: a box modeling approach II . PM 2 . 5 ,
coarse and PM 10 particulate matter fractions. vol. 36: p. 331–44, 2002.
[33] Schmitz R. Modelling of air pollution dispersion in Santiago de Chile. Atmos
Environ 2005;39(11):2035–47.
[34] Rabanal-Arabach J, Schneider A. Anti-reflective coated glass and its impact on bi-
facial modules’ temperature in desert locations. Energy Procedia 2016;92:590–9.
E.G. Luque et al.
[35] NREL, System Advisor Model Version 2016.3.14 (SAM 2016.3.14). 2016.
[36] Vasel A, Iakovidis F. The effect of wind direction on the performance of solar PV
plants. Energy Convers Manage 2017;153(August):455–61.
[37] Yusufoglu UA, Lee TH, Pletzer TM, Halm A, Koduvelikulathu LJ, Comparotto C,
et al. Simulation of energy production by bifacial modules with revision of ground
reflection. Energy Procedia 2014;55:389–95.
[38] Shoukry I, Libal J, Kopecek R, Wefringhaus E, Werner J. Modelling of bifacial gain
for stand-alone and in-field installed bifacial PV modules. Energy Procedia
2016;92:600–8.
625
Energy Conversion and Management 174 (2018) 615–625
[39] Kreinin L, Bordin N, Karsenty A, Drori A, Eisenberg N. Outdoor evaluation of power
output improvement of the bifacial module. In: Conference Record of the IEEE
Photovoltaic Specialists Conference, 2011.
[40] Faiman D, Berman D, Bukobza D, Kabalo S, Karki I, Medwedt B, et al. Field method
for determining the efficiency of each face of a Bi-facial photovoltaic module. World
Conf Photovolt Energy Convers 2003.
[41] Brennan MP, Abramase AL, Andrews RW, Pearce JM. Effects of spectral albedo on
solar photovoltaic devices. Sol Energy Mater Sol Cells 2014;124:111–6.