Summary of Research Paper

Part: 01,02

LECTURE # 46,47,48
 Solar Energy Materials & Solar Cells 95 (2011) 3359–3369

Contents lists available at SciVerse ScienceDirect

Solar Energy Materials & Solar Cells

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

A power-rating model for crystalline silicon PV modules

Thomas Huld a,n, Gabi Friesen b, Artur Skoczek c, Robert P. Kenny a, Tony Sample a,
Michael Field a, Ewan D. Dunlop a
a
European Commission, Joint Research Centre, Renewable Energy Unit, T.P. 450, Via E.l Fermi 2749, I-21020 Ispra (Va), Italy
b
University of Applied Sciences and Arts of Southern Switzerland, Institute of Applied Sustainability in the Built Environment, (SUPSI-ISAAC),
Via Trevano, CH-6952 Canobbio, Switzerland
c
GeoModel Solar s.r.o, Pionierska 15, 83102 Bratislava, Slovakia

a r t i c l e i n f o abstract

Article history: A model for the performance of generic crystalline silicon photovoltaic (PV) modules is proposed. The
Received 18 June 2010 model represents the output power of the module as a function of module temperature and in-plane
Received in revised form irradiance, with a number of coefficients to be determined by fitting to measured performance data
5 April 2011
from indoor or outdoor measurements. The model has been validated using data from 3 different
Accepted 25 July 2011
Available online 25 August 2011
modules characterized through extensive measurements in outdoor conditions over several seasons.
The model was then applied to indoor measurement data for 18 different PV modules to investigate the
Keywords: variability in modeled output from different module types. It was found that for a Central European
Crystalline silicon climate the modeled output of the 18 modules varies with a standard deviation (SD) of 1.22%, but that
PV energy rating
the between-module variation is higher at low irradiance (SD of 3.8%). The variability between modules
PV performance
of different types is thus smaller than the uncertainty normally found in the total solar irradiation per
Performance rating
year for a given site. We conclude that the model can therefore be used for generalized estimates of PV
performance with only a relatively small impact on the overall uncertainty of such estimates resulting
from different module types.
& 2011 Elsevier B.V. All rights reserved.

1. Introduction  the performance of PV modules changes (normally degrades)

with long-term exposure to outdoor conditions, which in turn
It is well known that the energy produced by a given photovoltaic affects the overall lifetime energy output (‘‘ageing effects’’).
(PV) system depends on several external factors. Foremost of these
factors is of course the amount of solar irradiation impinging on the The list is not exhaustive. Other influences on PV performance,
surface of the PV modules. However, other factors are of considerable such as soiling or snow cover, are not properties of the modules
importance: and will not be considered here.
The strength of these effects varies with PV technologies and
 PV power output depends on module temperature, which in even between modules using the same class of PV material. The
turn depends on ambient temperature and, the type of effects of module temperature and irradiance can be measured
mounting used, as well as wind speed and direction; indoors using a solar simulator to perform a set of power
 the power output is a nonlinear function of solar irradiance; measurements at different temperatures and irradiance levels to
 module surface reflectivity increases when the radiation hits produce a power matrix, which can be interpolated to a contin-
the module at an angle that differs from perpendicular uous performance surface. Alternatively, a short outdoor mea-
(‘‘angle-of-incidence effect’’); surement campaign is sufficient to provide enough data points to
 the efficiency of PV modules depends on the spectrum of the characterize the performance of the module [1–3]. A mathema-
sunlight, which in turn depends on sun height and meteor- tical model can then be fitted to these data points to yield a model
ological conditions (‘‘spectral effects’’); for the module power as a function of in-plane irradiance and
module temperature.
Various models developed for c-Si PV performance have been
compared [3–6], and have been shown to make fairly accurate
predictions of outdoor performance. However for a prospective
n
Corresponding author. Tel.: þ39 0332785273; fax: þ39 0332789268. buyer of PV modules this is not of great use, since there is no
E-mail address: Thomas.Huld@jrc.ec.europa.eu (T. Huld). standard model against which to compare individual module

0927-0248/$ - see front matter & 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.solmat.2011.07.026
3360 T. Huld et al. / Solar Energy Materials & Solar Cells 95 (2011) 3359–3369

Nomenclature N number of measurement data points used for the

modeling
Dij deviation of model for module i from measurement Nrad number of different radiation values for indoor
point j measurements
Zrel module efficiency relative to efficiency at STC Ntemp number of different temperature values for indoor
(dimensionless) measurements
G in-plane irradiance (W/m2) P(G,T) instantaneous module power (W)
GSTC in-plane irradiance at STC ( ¼1000 W/m2) PSTC,m module power, maximum power point at STC (W)
G0 normalized irradiance G¼G/1000 (dimensionless) STC standard test conditions, G ¼1000 W/m2, Tmod ¼25 1C,
k1,y,k6 coefficients of the module power model air mass¼1.5 global
k10 ,y,k60 coefficients of the module power model, scaled to a T temperature (1C)
nominal module power of 1 W Tmod module temperature (1C)
kT coefficient for module temperature as a function of Tamb ambient temperature (1C)
irradiance (1C m2 W  1) T0 module temperature relative to standard test condi-
tions, T 0 ¼ Tmod 25 (1C)

technologies for different climates, and often the nominal power curves were measured at temperatures ranging from 25 to
of the module is the only information the buyer has about a given 65 1C and different irradiance levels varying from 100 W/m2 to
module. 1000 W/m2. The Pasan LAPSS pulse shape is a fast, 3 ms long
For this reason, in this paper, we propose a model to predict rising pulse followed by a decaying tail decreasing exponentially
the energy output of a typical c-Si module with a reasonable for 40 ms. The lamp to test plane distance is about 7 m and the
accuracy for given input of irradiance and temperature. The aim irradiance uniformity is 71% in a 1.98 m diameter circle in the
of this paper is not to suggest a power model for c-Si that is test plane. Variable irradiance was obtained using the long
necessarily superior to all other models given in the literature but decaying tail of the Pasan light pulse. The test and the reference
to investigate whether such a model can be used to predict the device temperatures were recorded with Pt-100 temperature
power output of any c-Si module, and if so, with what sensors attached at their rear sides. Irradiance levels were
uncertainty. measured using a crystalline silicon reference cell [8,9]. The
If such a model can be shown to work it can be used by PV principle of this method is the determination of the I–V char-
simulation software, such as the web-based system PVGIS [7], acteristics of photovoltaic devices using a single flash solar
where the user can estimate the performance of a PV system of a simulator as light source and an appropriately sized active
given PV technology, at any location in the geographical areas electronic load. The electronic load imposes a rising voltage
covered, specifying only the nominal size of the system in kWp, (during the selected part of the light pulse) on the PV module.
and simple geometric data such as the inclination and orientation During this voltage sweep the values of voltage, current and
of the modules. For PVGIS, as well as for other PV simulation tools, irradiance are simultaneously recorded. From that data-triplet the
a general model for c-Si modules is useful for a number of full I–V curve can then be generated. The electronic load is
reasons: composed of a function generator driving a bipolar operational
power supply amplifier and a precision shunt. In addition a time
 data on existing module types are often not available to delay gate is used in order to delay the start of the rising voltage
construct a model for the given module type; until the irradiance has fallen to the required level.
 even if specific models could normally be constructed, new The temperature was varied by placing the module in a
module types will typically not immediately be in the module temperature controlled chamber, in which the circulating air
database of the simulation tool; was heated using electrical resistance heating elements, behind
 the user may have insufficient knowledge about the module a low-iron-content glass window.
type used. Since the irradiance varies during the I–V scan, special care
must be taken to recalculate the measured I–V curve to a constant
It should be emphasized that the present study only attempts irradiance level. In this study the correction procedure suggested
to account for the influences of irradiance and temperature. As in [10] has been employed.
mentioned above, there are other factors that influence the The measurement procedure has been subjected to an uncer-
overall energy production of a PV module. The relation of the tainty analysis [11]. According to this the uncertainty of the
present model to these effects will be discussed in Section 5. measurement of the maximum power point, expressed as 2 times
However, a detailed study of the variability of these effects for the standard deviation, is equal to 2.0%, corresponding to a
different crystalline silicon PV modules is outside the scope of standard deviation of 1.0%.
this study.

2.1.2. Spectral matching during indoor measurements

2. Experimental set-up In the method used to vary the irradiance levels for the
measurements, changes in the simulator spectrum will occur
2.1. Indoor measurements due to varying lamp filament temperature. Especially at low
irradiances this would strongly influence the measurement accu-
2.1.1. Measurement set-up racy for modules for which no spectrally matched reference
The performance of the modules was determined by measur- device is available. In the case of c-Si modules the mismatch
ing the I–V characteristics on a Large Area Pulsed Solar Simulator error is minimized by the use of a c-Si reference cell. For each
(LAPSS) following standard measurement procedures. Indoor I–V module type it has been verified that there is a linear relationship
 T. Huld et al. / Solar Energy Materials & Solar Cells 95 (2011) 3359–3369 3361

between the short-circuit current of the module and the irradi- The number of measurements varies between modules.
ance measured by the reference cell, indicating that the spectral Table 1 shows for each module the range and number of different
mismatch is small, following the procedures described in the IEC temperature and irradiance conditions made during the indoor
60904-10 standard [12]. characterization. For each module the measurements represent a
roughly rectangular grid in the G–T plane.

2.2. Outdoor long-term measurement set-up

3. Model
The outdoor characterization system has been described pre-
viously [13] and basically consists of a south-facing test rack, on
There exist a number of different mathematical models for the
which the modules are mounted in an open rack configuration.
performance of PV modules as a function of irradiance and
The module rack inclination may be fixed or changed with the
temperature, see for instance [15–21]. Friesen et al. [18] and
seasons, generally within the range 35–551. A number of modules
Dittmann et al. [19] have reported on intercomparisons of
may be connected and measured sequentially. Each module is
different models, including the one used in the present study,
connected to an appropriately sized bipolar operational power
and found that the models investigated performed well for
supply amplifier using a 4-wire connection, with a precision
crystalline silicon modules, provided that both the temperature
shunt resistor inserted on the negative line to measure the
and irradiance dependences were considered in the model. None
current. The system is computer controlled, enabling the power
of the models were clearly superior to the rest.
supplies to be swept from open circuit to short-circuit. The I–V
curve characteristics of each module were measured at regular
intervals of typically 4 min over the measurement period, the 3.1. Mathematical formulation
measurement of the I–V curve taking a few seconds for each
module on the rack. In between scans the module was connected The model proposed for the present is a variant of King0s
to an appropriately sized bipolar operational power supply model [15,16]. Variations on this model have been presented by a
amplifier and held close to the maximum power point to simulate number of authors [20–22]. The model applied in this work has
real operating conditions. The module temperature was measured the following form:
by a Pt-100 temperature sensor attached to the rear surface of the PðG0 ,T 0 Þ ¼ G0 ðPSTC,m þ k1 lnðG0 Þ þ k2 lnðG0 Þ2 þ k3 T 0 þ k4 T 0 lnðG0 Þ
module using thermal paste and aluminium tape. The tempera-
ture sensors were always placed close to the center of the module. þ k5 T 0 lnðG0 Þ2 þk6 T 02 Þ ð1Þ
The in-plane irradiance was measured with a calibrated c-Si where the normalized in-plane irradiance and module tempera-
reference cell. The c-Si reference cell is an ESTI sensor [14], and tures are given by
will be naturally closely matched to the c-Si module behavior
both in spectral and angle-of-incidence response as well as in G0  G=GSTC
response time to changing irradiance.
T 0  Tmod TSTC
The main difference from King0s original model is that the
2.3. PV modules used in the study
terms for current and voltage at maximum power point (MPP)
have been multiplied together to a single expression for the
A total of 18 modules have been used in the study. They are
module power at MPP. In this way the expression for the module
listed in Table 1. The first 8 modules are monocrystalline while
power is linear in the empirical coefficients PSTC,m and k1–k6, and
the other 10 use polycrystalline cells. The estimated power for
it is possible to fit the model to data that contain only the
each module in Table 1 is the value of PSTC,m as found by the fitting
measured power at given G and T.
procedure described in Section 3 below.
Another way of expressing this is in terms of the relative
conversion efficiency, defined as
Table 1
Range and number of measurements for each module, showing minimum and Zrel ðG0 ,T 0 Þ  PðG0 ,T 0 Þ=ðPSTC,m G0 Þ ð2Þ
maximum temperatures and irradiances, as well as the number of different
temperatures (Ntemp) and irradiances (Nrad) at which the measurements were The relative efficiency is the ratio of the module efficiency
performed. under given conditions of G and T to the efficiency at STC. The
Module PSTC,m Tmin Tmax Gmin Gmax Ntemp Nrad
empirical coefficients PSTC,m and k1–k6 must be found by fitting
number (W) (1C) (1C) (W/m2) (W/m2) the function to measured data (indoor or outdoor). The fit is done
by a least-square procedure. Although such a procedure exists in
1 46 25 60 50 1000 9 13 many software packages we have made use of the fact that Eq. (1)
2 135 23 65 175 1000 5 12
is linear in the coefficients, and developed a very fast algorithm
3 38 25 60 100 1000 4 6
4 225 25 45 100 1000 3 7 dedicated to performing the least-square fit. The least-square fit
5 144 25 60 100 1000 4 7 results in a set of 7 linear equations that has a unique solution.
6 165 25 60 100 1000 4 6 The description of the algorithm is not central to this investiga-
7 172 25 60 100 1000 4 6 tion, but may be consulted on the web page with supplementary
8 166 25 60 100 1000 4 6
9 55 25 60 50 1000 9 12
material for this paper: http://re.jrc.ec.europa.eu/esti/powermo
10 95 25 60 50 1000 9 13 del/index.html. The source code for the algorithm may also be
11 150 25 60 150 1000 4 11 downloaded there, together with the measurement data used for
12 173 24 60 125 1000 4 6 the fitting of the generic model.
13 200 25 60 100 1000 4 7
14 150 25 60 100 1000 4 7
15 165 25 60 100 1000 4 6 3.2. Procedure for the power-rating model for generic c-Si modules
16 222 25 60 100 1000 4 6
17 224 25 60 100 1000 4 6
18 152 25 60 100 1000 4 6
This study uses the indoor measurements of all modules as
described in part 2.
3362 T. Huld et al. / Solar Energy Materials & Solar Cells 95 (2011) 3359–3369

The procedure to generate and test a model for the power Table 2
output of a generic c-Si module is as follows: Quality of fitting procedure for each of the 18 modules and for the combined data
set, showing the relative mean bias error (RMBE) and standard deviation (RSD) for
the fit. N is the number of indoor measurements performed for each module.
1. for each module, use the (indoor) data set of irradiance,
temperature and maximum power to find the empirical con- Module PSTC (W) N MBE (%) SD (%)
stants of the model, PSTC,m and k1–k6 by least-square fitting;
2. to be able to directly compare modules with different nominal 1 45.6 117  0.00068 0.38
2 135.1 58  0.00155 0.179
power ratings, rescale the model coefficients by PSTC,m, that is 3 38.83 24  0.00049 0.054
0
PSTC,m ¼1, ki0 ¼ki/PSTC,m, i ¼1,y,6 for each module; 4 225.1 21  0.0076 0.161
3. for each module, rescale the original power values by dividing 5 143.9 28  0.0044 0.136
all measured power values by PSTC,m; 6 164.5 24  0.0024 0.095
7 172.2 24  0.00114 0.064
4. construct a new data set by pooling (rescaled) data points from
8 166.2 24  0.0023 0.072
all the modules; 9 54.56 105  0.0067 0.23
5. find an ‘‘average’’ model for the combined data set by fitting 10 95.16 115  0.0079 0.21
the model to the combined data set; 11 144.8 44  0.00099 0.068
6. apply the model to one or more outdoor data sets of in-plane 12 170 24  0.00159 0.081
13 197.5 28  0.0100 0.39
irradiance and module temperature to predict module output 14 152.2 28  0.0039 0.102
for that location, using first the ‘‘combined’’ fit coefficients and 15 165.8 24  0.0024 0.072
then each of the sets of fit coefficients for each module; 16 222 24  0.00112 0.059
7. compare the predicted values for each of the modules with 17 223.9 24  0.0034 0.127
18 152.5 23  0.00169 0.058
that of the ‘‘combined’’ model.
Combined 1 560  0.1 3.14

The indoor measurement data for the modules are not uni-
form; more measurements were performed for some of the
modules than for others. Adding all the data to the combined Table 3
data set would therefore give higher weight to the modules with RMBE and RSD taken over all modules for particular values of G and T. N is the
most measurement points. For this reason the largest individual number of measurements (from all modules) at the given conditions.

module data sets were pruned so that all the individual data sets Central G (W/m2) Central T (1C) RMBE(G,T) (%) RSD(G,T) (%) N
had a similar size. In particular, the data sets were generally
restricted to 4 temperature values: T ¼25, 35, 45 and 60 1C. 100 25  1.07 2.04 14
200 25  0.46 0.60 20
400 25 0.39 0.77 17
800 25 0.18 0.38 15
4. Model validation 100 60  1.87 2.66 13
200 60 0.16 0.49 20
4.1. Uncertainty of modeling procedure 400 60 0.25 0.71 17
800 60  0.07 0.27 16
For each of the modules in Table 1, the measured power matrix
was used to fit the model coefficients to that particular module.
The resulting power model for the ith module will in the STC power. Only for the fit to the combined data set is RSD higher,
following be denoted by Pi(G,T). For each of the modules the data at 3.14%. This is due to the larger spread of power values coming
were then rescaled to a nominal power PSTC ¼1 W, and the from measurements on different PV modules.
combined data set used to fit the model coefficients for an The values in Table 5 refer to the RMBE and RSD of all the
‘‘average’’ model (points 4 and 5 of the procedure described in measurements performed on a given module. However, the
Section 3.2). deviations might be more prevalent at particular irradiance and
For the jth measurement of module i, with measured irradi- temperature conditions. For instance, the model might consis-
ance, temperature and power Gij, Tij and Pij, respectively, we can tently predict too high or too low power at low irradiance. To
then calculate the relative deviation of the model from the investigate this, the RMBE and RSD were calculated over all
measurement at that point: modules for various combinations of G and T; that is
Dij ¼ ðPðGij ,Tij ÞPij Þ=Pij X
18  
RMBEðG,TÞ ¼ Pi ðG,TÞPij =Pij
The relative mean bias error of the fit for module i is then i¼1
given as
Ni
choosing only the instances of Pij where G and T lie in a narrow
1 X interval around the central values of G and T. SD(G,T) is calculated
RMBEi ¼ D ð3Þ
Ni j ¼ 1 ij in a similar way. In the following, the intervals are defined as
725 W/m2 and 75 1C.
and the relative standard deviation SDi as
Applying this analysis to 4 different irradiance values and 2
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P ffi
Ni
D2ij different temperatures, we obtain the results shown in Table 3.
j¼1
RSDi ¼ ð4Þ At very low irradiance (  100 W/m2) there is a negative bias of
Ni
about 1% at 25 1C, rising to nearly 2% at 60 1C. This shows that the
For each module, the RMBE and RSD of the fitting procedure model tends to underestimate (by  1%) the performance at low
are given in Table 2. irradiance. However, as will be shown in the next section, the
The mean bias error is very low, with an absolute value less total amount of energy that PV systems receive at this irradiance
than 0.01% of STC power for all the individual models, whereas is very slight so a 1% error at this irradiance has very small impact.
the RMBE of the combined model is about  0.1%. For the The case of G ¼100 W/m2 and T¼60 1C will almost never happen
individual model fits, RSD is consistently lower than 0.5% of the under outdoor conditions.
 T. Huld et al. / Solar Energy Materials & Solar Cells 95 (2011) 3359–3369 3363

Table 4 Table 6
RMBE and RSD for the fit to the pooled measurements for particular values of G Comparison of predicted and measured output from three modules measured
and T. N is the number of measurements (from all modules) at the given indoors and outdoors.
conditions.
Module Predicted Measured Deviation (%)
Central G (W/m2) Central T (1C) RMBE(G,T) (%) RSD(G,T) (%) N energy (MJ) energy (MJ)

100 25  1.31 5.45 14 Module 1 19.91 20.28  1.83

200 25  0.07 3.42 20 Module 9 59.69 59.78  0.15
400 25 0.53 3.7 17 Module 10 92.74 93.26  0.55
800 25  0.37 1.42 15
100 60  3.29 4.84 12
200 60 0.42 3.46 19 irradiance was measured with an ESTI sensor, which is an
400 60 0.29 2.5 16
encapsulated crystalline silicon reference cell [14]. In this way
800 60  0.26 1.28 14
the spectral and reflectivity characteristics will be very similar to
those of the modules [14], and therefore these effects will not be
compensated for explicitly.
Table 5
Power measurements were made only if the in-plane irradi-
Modules measured outdoors used for the model validation.
ance was larger than 50 W/m2.
Module 1 Module 9 Module 10 The model was applied for each module to the data sets
detailed in Table 5. The results are shown in Table 6, which
Peak power (W) 45.6 54.6 95.2 compares the modeled output with the measured energy produc-
Start date 2002-07-03 2003-01-01 2003-02-11
End date 2003-02-11 2003-12-31 2003-12-31
tion. The output is measured as the total energy output during the
Measurement 5 min 5 min (4 min from 5 min (4 min from 1-minute intervals when measurements were made, assuming
interval April) April) constant power during each one-minute interval. It is not
Inclination Summer, 351, Winter: 501, Winter: 501, intended as a measure of the total energy output during the
angles winter: 551 summer 351 summer 351
measurement campaign. This measure was chosen because of the
Number of 18,610 42,098 38,625
measurements varying time interval between measurements, which is anyway
rather irregular since measurements were not made at irradiance
levels o50 W/m2. The deviation of prediction from measured
The analysis in Table 3 concerns a comparison between each values is defined as
individual model fit Pi(G,T) to the measured data for that module. PN P
The same analysis can be performed for the combined fit, i¼1 PðGi ,Ti Þ N i ¼ 1 Pi
dev ¼ PN ð5Þ
comparing the combined fit P(G,T) to the measured data. The i¼1 i P
results are shown in Table 4.
The RMBE of this fit is similar to the individual fits to module From the results it is clear that 2 modules (Modules 9 and 10)
data, showing that the model has very little bias, except a value of have a very low difference between modeled and measured
around  1% at very low irradiance. In contrast the standard energy production. However, the 3rd module (Module 1) shows
deviation is considerably higher, indicating that there is a marked a considerably higher deviation ( 1.8% underprediction).
spread in measured power for different modules, and that this In order to investigate the model performance in more detail it
difference is higher than the measurement uncertainty and the was applied with each data set split into groups according to the
deviations of the fits from the measured data. This clearly shows in-plane irradiance. The results are shown in Fig. 1, with the
that the difference in the model output is due to differences in outdoor data split into 10 bins according to irradiance. For each
module performance and not a result of uncertainties in the bin the deviation between prediction and measurement is calcu-
modeling procedure. lated using Eq. (5).
Two prominent features can be seen in the deviation values. The
4.2. Comparison with outdoor measurements first is that the deviations for Module 1 are consistently negative, and
apart from the 50–100 W/m2 range the deviation is rather close to
Three of the modules in the study have also been measured the average deviation of –1.9%. This indicates that the relatively high
outdoors for extended periods (at least 6 months) at the JRC Ispra deviation for Module 1 is not due to a failure of the model to account
site (latitude 4514804300 N, longitude 81370400 E, elevation 220 m for the drop-off in performance with decreasing irradiance. A more
a.s.l.). For these modules the validity of the model was assessed probable explanation is a systematic error either in the indoor or the
by the following method: outdoor measurement system, leading to a small but noticeable
difference in power measurements.
 use indoor measurements to determine the coefficients of the When G 4100 W/m2 the model tracks the behavior of the
model by a least-square fit; module very well for Modules 1 and 9 whereas the model
 apply the model with the fitted coefficients to each outdoor significantly underestimates the performance of Module 10 when
measurement point using the measured in-plane irradiance G o300 W/m2.
and module temperature, in order to predict the power output The preceding analysis used the back-of-module temperature
for that measurement point; as input to the model. However, the back-of-module temperature
 compare the predicted energy output with the measured does not necessarily correspond to the temperature of the cell.
energy. King et al. [16] suggest a temperature difference between cell and
back-of-module that is proportional to irradiance. This will
The 3 modules measured outdoors at the JRC Ispra site are depend on the way the temperature sensor is mounted, on the
listed in Table 5 together with information about the measure- size of the module (the edges tend to be cooler than the center)
ments performed on each module. The peak power listed for the and on the wind speed. It is worth noting that if there is a
modules is the peak power estimated from the fitting procedure progressively increasing temperature discrepancy, this should
using the indoor data. For all outdoor measurements the show up in Fig. 1 as a steadily increasing difference between
3364 T. Huld et al. / Solar Energy Materials & Solar Cells 95 (2011) 3359–3369

Module 1
2 Module 9
Module 10

Deviation of prediction from measurement (%)

0

-2

-4

-6

-8

-10
0-100 1-200 2-300 3-400 4-500 5-600 6-700 7-800 8-900 >900
Irradiance ranges (W/m2)

Fig. 1. Percentage deviation of predicted output from measured output for each of the three modules measured both indoors and outdoors.

2
Module 1
Module 9
Module 10
Deviation of prediction from measurement (%)

-1

-2

-3
0-100 1-200 2-300 3-400 4-500 5-600 6-700 7-800 8-900 >900
Irradiance ranges (W/m2)

Fig. 2. Percentage deviation of predicted output from measured output for each of the three modules measured both indoors and outdoors, for a number of irradiance
ranges. In this case the prediction is based on the model fitted to a subset of the outdoor data for each module.

model and measurements. This could be the case for Module 10, In order to investigate whether the increasing discrepancy at
but does not seem to correspond to the results for Modules 1 and low irradiance is an inherent defect of the model or is due to
9. Since Modules 1 and 9 are rather small compared to Module 10, differences in the 2 measurement systems the same fitting
this may be due to edge effects. procedure has been employed, this time using a subset of the
If a cell temperature correction is applied according to [16], with outdoor data to perform the fit. For each of the 3 data sets 1% of
an assumed difference of 2 1C at G¼1000 W/m2, the overall effect will the data points were chosen by selecting every 100th data point.
be that the predicted energy output is reduced by  0.7% for all The model was then applied with these fitted coefficients to the
3 modules. In all 3 cases the discrepancy between model and entire outdoor data set. The results are shown in Fig. 2.
measurements will thus increase, although given the measurement In this case the overall deviations encountered are much lower,
uncertainties the discrepancy is really significant only for Module 1. though again with a higher deviation at the lowest irradiance levels.
 T. Huld et al. / Solar Energy Materials & Solar Cells 95 (2011) 3359–3369 3365

This may indicate a weakness of the model, although the uncertainty Table 7
of all measurements is also higher at this irradiance level. It should be Coefficients for the relative efficiency, obtained
from the fit to the combined data set of all
noted that under these measurement conditions (Ispra, Italy, tilted
modules.
surface), the range 50–100 W/m2 represents only 2% of the overall
energy arriving at the module. k01 (–)  0.01724
k02 (–)  0.04047
5. Results and discussion k03 (1C  1)  0.0047
k04 (1C  1) 1.49  10  4
k05 (1C  1) 1.47  110  4
The validation exercise has shown that the model is able to give a
k06 (1C  2) 5.0  110  6
good prediction of the power output for each of the modules used in
the investigation. It was also seen that the differences in module
performance are larger than the measurement and modeling uncer-
tainties and hence represent a real difference between the modules. It
is therefore possible to use the power models for each module to
estimate the difference in performance under realistic conditions, Table 8
Deviation in the predicted output from each module compared to the predicted
following points 6 and 7 of the procedure suggested in Section 3.2.
output of the model using the combined data set of all modules. This prediction
was made using a 1-yr data set of irradiance and module temperature measure-
5.1. Comparison of module performance for outdoor conditions ments from Ispra, Italy. The deviation was also calculated using two subsets of the
data set, one with only low irradiance values (G o 300 W/m2) and one with
medium to high irradiance (G 4300 W/m2).
A plot of the relative efficiency of the combined fit is shown in
Fig. 3. The coefficients k10 –k60 obtained from the combined data set Module number Deviation (%)

All data G 4300 W/m2 G o 300 W/m2

1.1 1  1.58  1.51  2.13

T=25deg. C
2 þ 0.53 þ 0.39 þ 1.64
T=60deg. C
3 þ 0.95 þ1.19  0.94
1 4 þ 0.08  0.20 þ 2.05
5  2.34  2.26  3.01
6 þ 0.67  0.21 þ 7.45
7  0.32  0.47 þ0.77
0.9 8  0.45  0.44  0.45
ηrel (-)

9  0.63  0.85 þ 1.06

10  0.82  0.13  6.16
11  1.81  1.18  6.58
0.8 12  0.43  0.35  1.05
13 þ2.58 þ2.42 þ 3.81
14  0.45  0.47  0.26
0.7 15  0.53  0.21  3.02
16  0.04  0.22 þ 1.40
17 þ 0.28 þ 0.41  0.69
18  2.18  1.40  8.20
0.6
100 200 300 400 500 600 700 800 900 1000
In-plane irradiance (W/m2)
are given in Table 7. In the figure, the error bars denote the
standard deviation of the measured data from the model predic-
1.1 tion for the values of G and T given in Table 4.
G=100W/m2
The 18 individual models and the combined model were then
G=800W/m2
applied to an outdoor data set consisting of in-plane irradiance
1 and module temperature. For this exercise, the outdoor data set
for Module 9 was chosen (see Section 4.2). For each model the
total energy production was calculated.
0.9 The output was then compared to the estimated output from
ηrel (-)

the model using the ‘‘combined’’ coefficients. The deviation in

output from the ‘‘combined’’ value is shown in Table 8.
0.8 The standard deviation of the estimates from the combined
estimate is 1.22%. This means that if the model is used with the
generic (or ‘‘combined’’) coefficients to estimate output for a
0.7 given module type there is a 90% probability that the model will
give a deviation that is not greater than about 2%, assuming a
normal distribution.
0.6 The value for the standard deviation is somewhat lower than
20 25 30 35 40 45 50 55 60 65 that reported in [2]. However, in that study, a comparison was
Tmod (C) made between predicted output from a model and real outdoor
measurements for the same modules. In this case only the
Fig. 3. (a) Cross sections of the efficiency surface for the combined model, as a differences in the predicted output between modules are con-
function of irradiance for two different module temperatures. (b) Cross sections of
the efficiency surface as a function of temperature for two different irradiances. In
sidered. This means that the uncertainty associated with the
both figures the error bars denote the standard deviation of the model from the outdoor measurements is not taken into account, including
measured data, considering the rescaled data from all modules. spectral effects, cover by soiling, snow and occasional shadows,
3366 T. Huld et al. / Solar Energy Materials & Solar Cells 95 (2011) 3359–3369

and initial module degradation. Therefore the spread in values 5.2. Variation in performance at low irradiance
between modules would be expected to be lower.
A graphical comparison of the performance of the various In order to investigate in more detail the source of the
modules is shown in Fig. 4. The figure shows the relative variation in performance between modules, the outdoor data set
efficiency as a function of irradiance for a fixed Tmod ¼40 1C and was divided into two parts: one consisting of all measurement
as a function of temperature for a fixed G ¼800 W/m2. To avoid points with G 4300 W/m2 and the complementary data set with
clutter in the graph only every 4th module is shown. It is clear Go300 W/m2. The models for each module and the combined
that most of the modules cluster around the curve of the model were then applied separately to the two data sets.
combined model, but a few are further away. Among the modules In the high irradiance case all the module predictions are close
shown in the graph are: Module 13, which gives the highest to the fit to the combined data set. The standard deviation is 1.1%,
estimated performance ( þ2.58% relative to the combined model), indicating that with a 90% confidence level (and assuming a
and Module 5, which gives the lowest estimated performance Gaussian distribution) the modules will perform within 1.7% of
(  2.34%). From the figures it is clear that Module 13 has a the average. For the low irradiance case the results are very
somewhat unusual behavior in that the highest efficiency is at different. In this case it was found that the standard deviation of
around G ¼500 W/m2, giving it a high relative efficiency. For the predictions is around 3.8%. This is clearly higher than the
Module 5 the relatively poor performance seems to be related variability using the full set of irradiance values
to the fact that the performance decrease with temperature is The fact that the low-irradiance performance is so much more
somewhat stronger than for the other modules. variable again shows that there really is higher variability between
modules in the low-irradiance end of the performance surface. For
the outdoor data set used in this analysis, periods of low irradiance
(o300 W/m2) account for about 12% of the total incoming solar
energy. The higher variability between modules applies only to this
fraction of the total incoming energy, which explains why the
1 overall variability between modules shown in Table 8 is not so
much higher than the variability found when considering only the
0.95 high irradiance fraction of the incoming energy. However, this
argument is valid only if the low-irradiance part of the irradiance
distribution is small. This may not be the case at high latitudes or in
0.9 cloudy climates and will be discussed in the next section.
The analysis so far has not distinguished between monocrys-
ηrel (-)

0.85 talline and polycrystalline PV modules. In the study, 8 out of 18

modules used monocrystalline cells. It is possible that while the
PV performance model works well for both types, there could be a
0.8 significant difference between the 2 classes of modules. This was
Combined model investigated by combining all the rescaled indoor data for each of
Module 1 the 2 module classes separately and applying the model fit to
0.75 Module 5
Module 9 each of the 2 data sets, in order to obtain separate models for
Module 13
Module 17 ‘‘generic’’ mono- and polycrystalline modules. When each of these
0.7 2 models are applied to the complete outdoor data set the
100 200 300 400 500 600 700 800 900 1000 monocrystalline model predicts 0.2% higher output than the
G (W/m2) combined model, while the polycrystalline model predicts 0.2%
lower output. This difference is much less than the spread in
values between individual modules (whether mono- or polycrys-
1.02
talline). Considering only irradiance above 300 W/m2 the differ-
1 ence is less than 0.1%. For the low-irradiance case G o300 W/m2
0.98 the combined monocrystalline model predicts 1.7% higher output
than the combined model for all modules, while the polycrystal-
0.96 line model predicts 1.3% lower output. This discrepancy will of
0.94 course be more marked in regions with a higher frequency of low-
irradiance conditions. The poorer low-light performance of some
0.92
ηrel (-)

polycrystalline modules can be related to the non-ideality of the

0.9 diode in the polycrystalline silicon material [23].
0.88
5.3. Geographical variation of model performance
0.86
Combined model
0.84
Module 1 The preceding analysis was made with a data set from a single
Module 5
Module 9 location, with a particular probability distribution of irradiance
0.82 Module 13
and temperature. Moving to a different location it can be expected
Module 17
0.8 that the average module efficiency will be affected by the local
25 30 35 40 45 50 55 60 conditions. In addition, the variation between modules may also
Module temperature (C) change especially due to changes in the frequency of occurrence
of low-light conditions.
Fig. 4. (a) Estimated relative efficiency as a function of irradiance for a constant To quantify this, the model was applied to data sets consisting of
module temperature Tmod ¼ 40 1C for a selection of modules and for the combined
model. (b) Estimated relative efficiency for the same modules, as a function of
time series of irradiance and ambient temperature for diverse
temperature for a constant G¼ 800 W/m2. Among the curves are those for the best locations. The irradiance data consist of hourly estimates of hor-
and worst performing modules (Modules 13 and 5, respectively). izontal global and diffuse irradiance from satellite images [24].
 T. Huld et al. / Solar Energy Materials & Solar Cells 95 (2011) 3359–3369 3367

Table 9 5.4.1. Angle-of-incidence effects

Annual average relative efficiency estimated by the combined model for five Direct measurements of the reflectivity as a function of
locations in Europe, together with the standard variation of the estimates for
incidence angle are not available for the modules that have been
individual modules.
used for the modeling described in Section 3. However, studies
Location Relative efficiency Standard deviation (%) found in the literature may provide some insight as to the
magnitude of the effect and how much it varies between modules.

Torshavn 0.940 2.64 Martin and Ruiz [28] described a model for the effects of shallow-
Hamburg 0.920 1.75
Gutsch 0.962 1.58
angle reflectivity, with separate expressions for the beam and
Ispra 0.899 1.16 diffuse irradiance. The model contains empirically determined
Albacete 0.897 0.97 coefficients, and the authors provide values for these coefficients
for a number of different PV module designs.
In order to investigate the magnitude and variation of this
The data set comprises the 11 yr period from 1996 to 2006. effect the model by Martin and Ruiz was applied to radiation data
Estimates of global and diffuse irradiance on an inclined plane were for a number of locations in Europe, using coefficients for
obtained using the model of Muneer [25] as implemented in the different module designs. The data used are the satellite-based
European Solar Radiation Atlas [26]. Temperature data were mea- radiation data described in Section 5.3. In all cases the mounting
sured by meterological stations at the sites chosen, with data is fixed south-facing with an inclination angle of 451. Fig. 5 shows
available at 3-hour intervals from 6:00GMT to 18:00GMT over the the estimated influence of angle-of-incidence for 3 different types
same time period as for the irradiance data. To obtain hourly values of module design, as described in [28], for 5 different locations in
of temperature a linear interpolation was employed. Europe. The results show that the annual losses due to angle-of-
The temperature measurements are of ambient temperature, incidence effects remain in the range of 2.5–4%, somewhat lower
so the module temperature must be estimated. For this study a than the effects of temperature and irradiance, given in Table 9
simple linear relationship between irradiance and module tem- (where relative efficiency ranging from 0.90 to 0.96 indicates that
perature has been employed [16,27]: losses due to irradiance and temperature range from 6% to 10%).
The geographical variation is also rather small, with a range of
Tmod ¼ Tamb þ kT G
only 1.5%. On the other hand the variation between different
The coefficient kT may be found by fitting to a data set measured technologies is around 0.8%, which is somewhat smaller but not
outdoors for a given module. Typical values for free-standing dramatically different from the variation in the effect of tempera-
rack-mounted modules is kT ¼0.03–0.035 1C m2 W  1. For the ture and irradiance between module types in this study.
present study, a value of kT ¼0.035 1C m2 W  1 was chosen. This For a given geographical location the angle-of-incidence effect
does not take into account the cooling effect of wind, as no wind will depend on the module inclination angle, which again may be
data were available, so the estimated values of the module fixed or variable (for tracking systems). For a discussion of the
efficiency may be somewhat different for windy locations. How- influence of mounting type, see [29].
ever the general trend remains, and the variation between Since the effect of reflectivity prevents some of the radiation
modules should be similar, since this seems to be caused mainly from reaching the PV material in the modules, the irradiance should
by differences in the low-irradiance performance. be corrected for this effect before applying it to the power model.
Using the method described above, the data sets were applied
to the power model for each of the 18 modules and for the
5.4.2. Spectral effects
combined model. For each location the average annual relative
The sensitivity of crystalline silicon PV cells to different light
efficiency was found. From the values for the 18 individual
wavelengths (the ‘‘spectral response’’) is well known (see for
modules the standard deviation from the value of the combined
model can then be found, using Eq. (5). The results are shown in
4.5

Table 9 for 5 locations: Torshavn (62110N 61460W), Hamburg Glass-SiO2-Si
Glass-ZnS-Si
(531380N 91590E), Gutsch (461390N 81370E), Ispra (451480N 4 Glass-triplecoating-Si
81370E) and Albacete (381570N 11510W). The station at Gutsch is
at elevation 2284 m a.s.l, whereas the other locations are at low 3.5
altitude.
Annual energy loss (%)

The results clearly show a decreasing trend in the variation in 3

(simulated) output between modules as the system moves from
northerly locations with cloudy climate towards more sunny 2.5
climatic conditions in the south. The higher uncertainty in more
cloudy areas should be taken into account when assessing the 2
quality of PV performance simulations for different locations.
1.5
5.4. Role of power model in relation to other effects
1
Temperature and irradiance are not the only physical effects
0.5
influencing the performance of PV modules, as described in the
Introduction section.
0
The model proposed in the previous sections does not take into Torshavn Hamburg Gutsch Ispra Albacete
account these effects, and a detailed investigation is outside the
scope of the present paper. What follows is only a brief discussion Fig. 5. Yearly loss of irradiation due to shallow-angle reflectivity, for five different

locations in Europe: Torshavn (62110N 61460W), Hamburg (531380N 91590E), Gutsch
of the magnitude of these effects for the overall PV system
(461390N 81370E), Ispra (451480N 81370E) and Albacete (381570N 11510W). The
performance, in order to compare the importance of these effects losses are given as a percentage of the in-plane irradiation on a south-facing plane
with that of the effects (temperature and irradiance) included in inclined at 451. Three different PV designs are considered, all with crystalline
the power model. silicon cells.
3368 T. Huld et al. / Solar Energy Materials & Solar Cells 95 (2011) 3359–3369

instance Żdanowicz et al. [30] or Kenny et al. [1]). Unfortunately, or 20% for the lifetime of the modules (20 yr). This uncertainty is
the spectrum of the radiation impinging on PV modules under much higher than the variation between modules that was found
realistic conditions is more uncertain. For clear-sky conditions the for the effects of temperature and irradiance. Dunlop [37], in a
spectrum is reasonably well known as a function of air mass, similar work, demonstrated for some 144 different silicon photo-
pressure and relative humidity [31]. However, for turbid cloudless voltaic modules that the average degradation on initial outdoor
skies and for cloudy/overcast conditions, the effects on the exposure is 2.6%. If this is assumed in the above case then after
spectrum are more complex. Gottschalg et al. [32] suggested the initial degradation (referred to as photon degradation) the
calculating the effect of changing spectrum by means of the average annual degradation would then be approximately 0.7%
Useful Fraction (UF) of the solar radiation, i.e. the fraction of the per annum.
solar radiation that is usable for a given PV technology. These It should be emphasized that the estimate of performance
authors estimated UF as a function of clearness index and air degradation depends on a number of factors that are not well
mass. In later publications [33,34] the authors suggested the use known. The estimate is based only on measurements at STC, so it
of the average photon energy, which however requires knowledge is not known if the performance dependence on temperature;
of the spectrum under varying meteorological conditions at the irradiance and spectrum also change. It is also not known if the
given location. degradation occurs linearly over time or if a significant part of the
The magnitude and even the sign of the spectral effect on c-Si degradation happens near the beginning of the exposure period. If
are not obvious. For clear-sky conditions different authors have this is the case the lifetime performance loss will be greater.
reported conflicting results. King et al. [16] and Żdanowicz et al. Finally such studies are necessarily based on PV modules pro-
[30] report an increased efficiency of c-Si at high air mass and duced long ago, mainly in the 1980s. Manufacturing technology
clear-sky conditions. On the other hand Kenny et al. [1] show a has changed significantly since then, making it uncertain whether
slight decrease in efficiency with increasing air mass for a c-Si the conclusions drawn from studies such as this are still appro-
module mounted on a tracker under clear-sky conditions. priate to current module technologies.
Huld et al. [35] estimated the effects of varying spectral In terms of combining the estimate of the degradation with the
content on the performance of PV modules, by looking at the other models discussed here, existing results do not make it
effect of differing weather conditions on the short-circuit current possible to make modifications to the power model itself. If we
of modules measured outdoors over an extended time period. It are to estimate the module energy life (over 20 yr) and not the
was found that the overall effect for a crystalline silicon module energy rating as discussed in this paper the most appropriate way
was an increase in output of 0.7% over a whole year, relative to would be to use the model proposed here but de-rate the module
the performance at air mass 1.5 global (the air mass to be used as power at STC for each year of calculation.
Standard Test Condition). This result is valid only for a Central
European climate, but the magnitude of the effect is so small as to
nearly vanish in the measurement uncertainties. The effect is 6. Conclusions
much smaller than the effect of temperature and irradiance on
c-Si modules, and smaller even than the variation between c-Si In this paper a model has been presented for the performance
modules. Incorporating a model for spectral variations will there- of c-Si modules representing the power output as a function of in-
fore only lead to a slight change in overall estimated energy plane irradiance and module temperature. The model is based on
output, compared with the variability of PV module performance a series of power measurements performed indoors, used to
and the uncertainties in the estimates of solar irradiation, which determine the empirical coefficients of the model. The model
may be of the order of several percent. has been found to fit the indoor measurement matrix with
negligible bias except at very low irradiance, where the deviation
may be up to  1%. The model has also been validated by applying
5.4.3. Module degradation due to ageing the model to 3 long-term (6–12 months) outdoor data sets and
The PV modules studied in this work were all relatively new at comparing with the actual power values measured during that
the time of measurement, and none of them had been exposed to time. For 2 modules the prediction was well within 1% of the
long-term outdoor conditions at the time of the indoor measure- actual energy output of the module, and predicted well the
ments. Therefore the predicted module energy output or rating performance of the model over a wide range of irradiances above
from the method given here is applicable only to the case of 100 W/m2; below this level the model predictions were signifi-
newly installed modules. After installation all photovoltaic mod- cantly lower (5–10%) than measurements. For the 3rd module the
ules are known to decrease in their power and therefore energy model predicted about 2% less energy, a value that was approxi-
generation over the following years. mately constant except at irradiance (o200 W/m2), indicating
Skoczek et al. [36] have reported results of measurements that the discrepancy may be due to problems with the indoor
performed on a number of PV modules exposed outdoors at the measurements.
JRC Ispra site for an extended period of time, generally more than The model was then used to fit indoor measurement data for
20 yr. The modules were not measured continuously but mainly 18 different c-Si modules to provide a specific model for each
at the beginning and end of exposure. The mean calculated power module and to create a ‘‘generic’’ c-Si module model from them.
loss of all 204 modules was 17.3% over 21 yr. It is not known These models were then applied to an outdoor data set of
precisely how the degradation proceeds in time and in order to irradiance and module temperature values and the simulated
estimate the overall impact on PV performance some simplifying energy output compared. The results were found to be very
assumptions must be made. similar, with the individual module predictions clustering around
For all 204 modules examined it was found that the average the prediction from the generic model with a standard deviation
annual decrease in performance over 21 yr is approximately 0.8% of 1.22%. Considering only data with G 4300 W/m2 the agree-
(excluding modules that have broken down altogether). This ment was even better, with a SD of 1.0%.
value is comparable to the loss of output due to temperature Analyzing the mono- and polycrystalline modules separately it
and irradiance effects. However, the modules investigated varied was found that the difference between the two groups and the
strongly in the degradation of performance. The standard devia- generic model is less than 0.2%, which is significantly smaller than
tion of the annual performance degradation is approximately 1%, the deviation between any two arbitrary modules. Thus treating
 T. Huld et al. / Solar Energy Materials & Solar Cells 95 (2011) 3359–3369 3369

mono- and polycrystalline silicon modules as part of a single [13] D. Anderson, J. Bishop, E. Dunlop, Energy rating of photovoltaic modules,
group is a reasonable approximation. in: Proceedings of the 16th European Photovoltaic Solar Energy Conference.
Glasgow, UK, May 2000, pp. 2087–2091.
This uncertainty of 2% found in this study is considerably [14] C. Helmke, E.D. Dunlop, K.T. O’Hara, H.A. Ossenbrink, Analyses of optical
smaller than the uncertainty in irradiation values that can be artefacts observed comparing the ESTI-sensor and a pyranometer, in:
obtained for an arbitrary location in Europe, where different Proceedings of the 14th European Photovoltaic Solar Energy Conference,
Barcelona, Spain, 1997, pp. 1640–1643.
information sources may disagree by 10% or more. This means [15] D.L. King, J.A. Kratochvil, W.E. Boyson, W.I. Bower, Field experience with a
the assumption that all c-Si modules have substantially similar new performance characterization procedure for photovoltaic arrays, in:
behavior will not significantly increase the uncertainty in PV Proceedings of the second World Conference and Exhibition on Photovoltaic
Solar Energy Conversion, Vienna, 1998, pp. 1947–1952.
output predictions. Thus, the model as presented here provides a [16] D.L. King, W.E. Boyson, J.A. Kratochvil, Photovoltaic Array Performance
good estimate for the actual energy output of an arbitrary c-Si PV Model, SAND2004-3535, Sandia National Laboratories, 2004.
system. [17] E. Skoplaki, J.A. Palyvos, On the temperature dependence of photovoltaic
module electrical performance: a review of efficiency/power correlations,
The model as implemented here has been included in the
Solar Energy 83 (2009) 614–624.
online PV estimator in PVGIS (http://re.jrc.ec.europa.eu/pvgis/ [18] G. Friesen, S. Dittmann, S.R. Williams, R. Gottschalg, H.G. Beyer, A. Guerin de
apps4/pvest.php) and will be updated as soon as a higher number Montgareuil, N. Van der Borg, A.R. Burgers, R. Kenny, T. Huld, B. Müller,
of indoor measurements of different c-Si module types will be C. Reise, J. Kurnik, M. Topić, Intercomparison of different energy prediction
methods within the European project ‘Performance’—results of the 2nd
available. round robin, in: Proceedings of the 24th European Photovoltaic Solar Energy
The results of this study are applicable only to c-Si modules. Conference (2009) 3189–3197.
Other module technologies (including hybrid technologies such as [19] S. Dittmann, G. Friesen, S. Williams, T.R. Betts, R. Gottschalg, H.G. Beyer,
A. Guérin de Montgareuil, N..v.d. Borg, A.R. Burgers, T. Huld, B. Müller, C.
a-Si/c-Si) will need a similar analysis in order to provide generic Reise, J. Kurnik, M. Topić, T. Żdanowicz, F. Fabero, Results of the 3rd modeling
models for these technologies. round robin within the European project PERFORMANCE,—comparison of
The module measurements and the modeling software are module energy rating methods, in: Proceedings of the 25th European
Photovoltaic Solar Energy Conference, 2010, pp. 4333–4338.
available for download at http://re.jrc.ec.europa.eu/esti/generic_c [20] D. Chianese, G. Friesen, N. Cereghetti, A. Realini, E. Bura, S. Rezzonico,
si_model.html. A. Bernasconi, Final Report 2000–2003: qualita e resa energetica di moduli
ed impianti fotovoltaici—centrale LEEE-TISO; SFOE Contract No. 36508.
[21] R.P. Kenny, G. Friesen, D. Chianese, A. Bernasconi, E.D. Dunlop, Energy rating
of PV modules: comparison of methods and approach, in: Proceedings of the
Acknowledgments third World Conference on Photovoltaic Energy Conversion, Osaka, Japan,
2003, pp. 2015–2018.
[22] T. Huld, M. Šúri, E.D. Dunlop, Geographical variation of the conversion
This work was partially funded by the PERFORMANCE Project efficiency of crystalline silicon photovoltaic modules in Europe, Progress in
of the European Commission under Contract number SES-019718 Photovoltaics: Research and Applications 16 (2008) 585–607.
[23] D. Anderson, G. Agostinelli, T. Sample, E. Dunlop, Modeling outdoor behavior
within FP6.
of fill factor in thin film modules, in: Proceedings of the 16th European
Photovoltaic Solar Energy Conference, Munich, Germany, 2001, pp. 470–473.
[24] A. Hammer, D. Heinemann, A. Westerhellweg, P. Ineichen, J.A. Olseth,
References A. Skartveit, L. Wald, H.G. Beyer, C. Reise, Derivation of daylight and solar
irradiance data from satellite observations, in: Proceedings of the ninth
Conference on Satellite Meteorology and Oceanography, Paris, France, 1998,
[1] R.P. Kenny, A. Ioannides, H. Müllejans, W. Zaaiman, E.D. Dunlop, Performance
pp. 747–750.
of thin film PV modules, Thin Solid Films 511–512 (2006) 663–672. [25] T. Muneer, Solar radiation model for Europe, Building Services Engineering
[2] G. Friesen, D. Chianese, I. Pola, A. Realini, A. Bernasconi, Energy rating Research and Technology 11 (1990) 153–163.
measurements and predictions at ISAAC, in: Proceedings of the 22nd [26] K. Scharmer, J. Greif (Eds.), The European Solar Radiation Atlas, vol. 2,
European Photovoltaic Solar Energy Conference, Milan, Italy, 2007, Database and exploitation software, Presses de l’École des Mines, Paris, 2000.
pp. 2754–2757. [27] E. Skoplaki, A.G. Boudouvis, J.A. Palyvos, A simple correlation for the
[3] G. Friesen, R. Gottschalg, H.G. Beyer, S. Williams, A. Guérin de Montgareuil, operating temperature of photovoltaic modules of arbitrary mounting, Solar
T. Huld, B. Müller, A.C. de Keizer, Y. Niu, Intercomparison of different energy Energy Materials and Solar Cells 92 (2008) 1393–1402.
prediction methods within the European project ‘Performance’. Results of the [28] N. Martin, J.M. Ruiz, Calculation of the PV modules angular losses under field
1st round robin, in: Proceedings of the 22nd European Photovoltaic Solar conditions by means of an analytical model, Solar Energy Material & Solar
Energy Conference, Milan, Italy, 2007, pp. 2659–2663. Cells 70 (2001) 25–38.
[4] A. Guérin de Montgareuil, Description of MOTHERPV, the new method [29] T. Huld, M. Šuri, E.D. Dunlop, Comparison of potential solar electricity output
developed at INES/CEA for the assessment of the energy production of from fixed-inclined and two-axis tracking photovoltaic modules in Europe,
photovoltaic modules, in: Proceedings of the 22nd European Photovoltaic Progress in Photovoltaics: Research and Applications 16 (2008) 47–59.
Solar Energy Conference, Milan, Italy, 2007, pp. 2608–2612. [30] T. Żdanowicz, T. Rodziewicz, M. Za˛browska-Wac"awek, Effect of airmass
[5] B. Marion, Comparison of predictive models for photovoltaic module perfor- factor on the performance of different type of PV modules, Opto-Electronics
mance, in: Proceedings of the 33rd IEEE Photovoltaic Specialists Conference, Review 12 (1) (2004) 69–73.
San Diego, California, May 11–16, 2008, pp. 1–6. [31] C. Gueymard, D. Myers, K. Emery, Proposed reference irradiance spectra for
[6] W. Heidenreich, B. Müller, C. Reise, Describing the world with three para- solar energy systems testing, Solar Energy 73 (2002) 443–467.
meters: a new approach to PV module power modeling, in: Proceedings of [32] R. Gottschalg, D.G. Infield, M.J. Kearney, Experimental study of variations of
the 23rd European Photovoltaic Solar Energy Conference, Valencia, Spain, the solar spectrum of relevance to thin-film solar cells, Solar Energy 79
2008, pp. 2786–2789. (2003) 527–537.
[7] M. Šúri, T. Huld, E.D. Dunlop, PV-GIS: a web based solar radiation database for [33] R. Gottschalg, T.R. Betts, S.R. Williams, D. Sauter, D.G. Infield, M.J. Kearney,
the calculation of PV potential in Europe, International Journal of Sustainable A critical appraisal of the factors affecting energy production from amor-
Energy 24 (2) (2005) 55–67. phous silicon photovoltaic arrays in a maritime climate, Solar Energy 77
[8] IEC Central Office: Measurement of Photovoltaic Current–voltage Character- (2004) 909–916.
istics, IEC-60904-1 s edition, 2006. [34] R. Gottschalg, T.R. Betts, D.G. Infield, M.J. Kearney, The effect of spectral
[9] IEC Central Office Photovoltaic devices—Procedures for Temperature and variations on the performance parameters of single and double junction
Irradiance Corrections to Measured I–V Characteristics, IEC-60891, second amorphous silicon solar cells, Solar Energy Materials and Solar Cells 85
edition, 2009. (2005) 415–428.
[10] H. Müllejans, D. Pavanello, A. Virtuani, New correction procedure for I–V [35] T. Huld, T. Sample, E.D. Dunlop, A simple model for estimating the influence
curves measured under varying irradiance, in: Proceedings of the 23rd of spectrum variations on PV performance, in: Proceedings of the 24th
European Photovoltaic Solar Energy Conference, Valencia, Spain, 2008, European Photovoltaic Solar Energy Conference, Hamburg, Germany, 2009,
pp. 2818–2821. pp. 3385–3389.
[11] H. Müllejans, W. Zaaiman, R. Galleano, Analysis and mitigation of measurement [36] A. Skoczek, T. Sample, E.D. Dunlop, The results of performance measurements
uncertainties in the traceability chain for the calibration of photovoltaic devices, of field-aged crystalline silicon photovoltaic modules, Progress in Photovol-
Measurement Science and Technology (2009). doi:10.1088/0957-0233/20/ taics: Research and Applications 17 (2009) 227–240.
075101. [37] E.D. Dunlop, Lifetime performance of crystalline silicon PV modules,
[12] IEC Central Office, Methods of Linearity Measurement IEC-60904-10 s in: Proceedings of the third World Conference on Photovoltaic Energy
edition, 2009. Conversion, Osaka, Japan, 2003, pp. 2927–2930.
 How to Summarize a Research Article

Research articles use a standard format to clearly communicate information about an

experiment. A research article usually has seven major sections: Title, Abstract,
Introduction, Method, Results, Discussion, and References.

Determine your focus

The first thing you should do is to decide why you need to summarize the article. If the
purpose of the summary is to take notes to later remind yourself about the article you may
want to write a longer summary. However, if the purpose of summarizing the article is to
include it in a paper you are writing, the summary should focus on how the articles
relates specifically to your paper.

Reading the Article

Allow enough time. Before you can write about the research, you have to understand it.
This can often take a lot longer than most people realize. Only when you can clearly
explain the study in your own words to someone who hasn’t read the article are you ready
to write about it.

Scan the article first. If you try to read a new article from start to finish, you'll get
bogged down in detail. Instead, use your knowledge of APA format to find the main
points. Briefly look at each section to identify:
• the research question and reason for the study (stated in the Introduction)
• the hypothesis or hypotheses tested (Introduction)
• how the hypothesis was tested (Method)
• the findings (Results, including tables and figures)
• how the findings were interpreted (Discussion)

Underline key sentences or write the key point (e.g., hypothesis, design) of each
paragraph in the margin. Although the abstract can help you to identify the main points,
you cannot rely on it exclusively, because it contains very condensed information.
Remember to focus on the parts of the article that are most relevant.

Read for depth, read interactively. After you have highlighted the main points, read each
section several times. As you read, ask yourself these questions:
• How does the design of the study address the research questions?
• How convincing are the results? Are any of the results surprising?
• What does this study contribute toward answering the original question?
• What aspects of the original question remain unanswered?

Plagiarism. Plagiarism is always a risk when summarizing someone else’s work. To

avoid it:
• Take notes in your own words. Using short notes or summarizing key points in
your own words forces you to rewrite the ideas into your own words later.
• If you find yourself sticking closely to the original language and making only
minor changes to the wording, then you probably don't understand the study
Writing the Summary
Like an abstract in a published research article, the purpose of an article summary is to
give the reader a brief overview of the study. To write a good summary, identify what
information is important and condense that information for your reader. The better you
understand a subject, the easier it is to explain it thoroughly and briefly.

Write a first draft. Use the same order as in the article itself. Adjust the length
accordingly depending on the content of your particular article and how you will be using
the summary.
• State the research question and explain why it is interesting.
• State the hypotheses tested.
• Briefly describe the methods (design, participants, materials, procedure, what was
manipulated [independent variables], what was measured [dependent variables],
how data were analyzed.
• Describe the results. Were they significant?
• Explain the key implications of the results. Avoid overstating the importance of
the findings.
• The results, and the interpretation of the results, should relate directly to the
hypothesis.

For the first draft, focus on content, not length (it will probably be too long). Condense
later as needed. Try writing about the hypotheses, methods and results first, then about
the introduction and discussion last. If you have trouble on one section, leave it for a
while and try another.

If you are summarizing an article to include in a paper you are writing it may be
sufficient to describe only the results if you give the reader context to understand those
results.

For example: “Smith (2004) found that participants in the motivation group scored higher
than those in the control group, confirming that motivational factors play a role in
impression formation”. This summary not only tells the results but also gives some
information on what variables were examined and the outcome of interest. In this case it
is very important to introduce the study in a way that the brief summary makes sense in
the larger context

Edit for completeness and accuracy. Add information for completeness where necessary.
More commonly, if you understand the article, you will need to cut redundant or less
important information.

Stay focused on the research question, be concise, and avoid generalities.

Edit for style. Write to an intelligent, interested, naive, and slightly lazy audience (e.g.,
yourself, your classmates). Expect your readers to be interested, but don't make them
struggle to understand you. Include all the important details; don't assume that they are
already understood.
• Eliminate wordiness, including most adverbs ("very", "clearly"). "The results
clearly showed that there was no difference between the groups” can be shortened
to "There was no significant difference between the groups".
• Use specific, concrete language. Use precise language and cite specific examples
to support assertions. Avoid vague references (e.g. "this illustrates" should be
"this result illustrates").
• Use scientifically accurate language. For example, you cannot "prove"
hypotheses (especially with just one study). You "support" or "fail to find support
for" them.
• Rely primarily on paraphrasing, not direct quotes. Direct quotes are seldom
used in scientific writing. Instead, paraphrase what you have read. To give due
credit for information that you paraphrase, cite the author's last name and the year
of the study (Smith, 1982).
• Re-read what you have written. Ask others to read it to catch things that you’ve
missed.

Adapted from: Summarizing a Research Article 1997-2006, University of Washington