Proceedings Chapter

Performance assessment of a simulation model for PV modules of

any available technology

MERMOUD, Andr, LEJEUNE, Thibault

Abstract

From long-term detailed measurements of several PV modules of all commercialized

technologies, this work aims to analyze the results of the standard one-diode model, and
suggests some modifications for improving it, especially for amorphous, microcrystalline and
CdTe modules. We found that for any module an exponential behaviour of the shunt
resistance parameter should be taken into account. We identified two other corrections
(recombination losses and spectral correction) in order to improve the modelling of
amorphous technology modules. These improvements have been implemented in the PVsyst
software developed at the University of Geneva.

Reference
MERMOUD, Andr, LEJEUNE, Thibault. Performance assessment of a simulation model for PV
modules of any available technology. In: Proceedings of the 25th European Photovoltaic
Solar Energy Conference. Mnchen : WIP, 2010.

Available at:
http://archive-ouverte.unige.ch/unige:38547

Disclaimer: layout of this document may differ from the published version.
 25th European Photovoltaic Solar Energy Conference Valencia, Spain, 6-10 September 2010

PERFORMANCE ASSESSMENT OF A SIMULATION MODEL FOR PV MODULES OF ANY AVAILABLE

TECHNOLOGY

Andr Mermoud, Thibault Lejeune

Institute for Environmental Sciences / Energy Group, University of Geneva,
Battelle, Bt. D, 7, route de Drize, CH-1227 Carouge Switzerland
[email protected] [email protected]

ABSTRACT: From long-term detailed measurements of several PV modules of all commercialized technologies, this
work aims to analyze the results of the standard one-diode model, and suggests some modifications for improving it,
especially for amorphous, microcrystalline and CdTe modules. We found that for any module an exponential behaviour
of the shunt resistance parameter should be taken into account. We identified two other corrections (recombination losses
and spectral correction) in order to improve the modelling of amorphous technology modules. These improvements have
been implemented in the PVsyst software developed at the University of Geneva.
Keywords: Modelling, Performance, Thin film.

1 INTRODUCTION selection process mainly according to irradiance stabili-

ty our sample contains several thousand of records for
Thin film photovoltaic (PV) modules (amorphous a- each module.
si:H, CIS, CdTe or C-a:Si technologies) present the
most promising opportunity to significantly decrease the
prices of PV in the future, as they need a very limited 3 ESTABLISHING THE STANDARD MODEL
quantity of pure materials, and the manufacturing
processes may be simplified. Their energetic return time A valid model should reproduce the electrical beha-
is usually lower than one year. viour of a PV module under any external conditions
Their commercial development is rapidly growing. (irradiance, temperature, incidence angle, spectral con-
Therefore any simulation tool should be able to accurate- tents). Furthermore, in order to be applied in a simula-
ly evaluate their performances in a PV system in real tion software, the model should be established with a
conditions, and to compare them with the traditional Si- minimum number of extra parameters not usually
crystalline solution provided by the present-day manufacturers datasheets.
Many teams report measurements of whole PV sys-
The starting point of this study is the usual one-diode
tems equipped with thin film modules, and observe sea-
model (hereafter referred as the standard model), estab-
sonal, irradiance or temperature behaviours different
lished for a crystalline-Si cell and extended to the whole
from crystalline modules. But there is no consensus on
module (Duffie et al. [2]):
how to interpret these data and a fortiori how to model
them.


" % +  '(
 =      ! 1$
The aim of this study is to establish a consistent mod-
')*
(1)
el that can be used in software dedicated to PV systems
simulation (like PVsyst).
where
, %: Current and voltage at modules terminals,
, : Photocurrent at the measured irradiance -./0( , pro-
2 MEASUREMENTS
portional to irradiance (1,2/3 at the reference
Our approach is mainly phenomenological, based on diance -2/3 )
detailed comparisons of outdoor measurements of the IV  : Diode saturation current, varies exponentially with
characteristics and model predictions. Our PV module temperature, (,2/3 at the reference temperature 42/3 ).
test facility is located on the roof of our building at the ')* : Shunt resistance, inverse of the slope around short-
Geneva University, and operates since 2004 (Mermoud circuit ((5 ) point,
[4]). Outdoor measurements are recorded every 10 mi- ') : Series resistance (may vary between 0 and a ')678
nutes in order to provide a significant sample containing value),
all irradiance and temperature conditions. These mea- 9: Diode ideality factor (should normally lie between
surements are performed on 8 modules simultaneously 1 and 2 per junction),
one mono-crystalline for calibrations, one CIS, and others :: Charge of the electron,
of various technologies. ;<) : Number of cells in series
Each record includes 30 points distributed along the
IV curve and measurements of irradiances with pyrano-
For the generalization to other technologies, we try to
meters (global in the module plane, Glob , global and
identify the modifications required to match the measured
diffuse in the horizontal plane, Glob
and Diff
respec-
data (cf. Mermoud [4] for a complete description of our
approach). The final objective is to describe the module
tively), as well as the modules and external tempera- behaviour under any operating conditions, from a unique
tures. In addition, irradiances measured by a PV-cell are set of parameters.
recorded before, at the middle and at the end of the mea- The 5 basic parameters (1,2/3 , ,2/3 , ')* , ') ,
9) are established using one I/V characteristics, chosen in
surement in order to check the stability of the irradiance
(the pyranometer has a time constant of 30 sec). After a
 25th European Photovoltaic Solar Energy Conference Valencia, Spain, 6-10 September 2010

our sample as a reference curve. For the corresponding U0[/2 they should be renormalized by U\?. /U0[/2 , with
reference conditions (-2/3 , 42/3 ), we can directly estab- U0[/2 usually about half of U\?. .
lish the '(* parameter (slope around (5 ), and write the When using the model in a simulation software, the
equation (1) at 3 points ()< , 0), (%.> , .> ), (0, %?5 ). Then, ()< , %V< , %6W , 6W ) parameter set will be the manufac-
setting the ') parameter, we can solve the equations in turers specified STC values. As they will not be exactly
order to determine 1,2/3 , ,2/3 , and 9. We adjust ') representative of your module, you may have significant
with the value that provides the best match of I/V curve. errors. But these errors are related to the parameters
This provides the model for expressing the full I/V uncertainties, not to the model quality. The MBD can be
characteristics at reference conditions. significant, but if the model is good (with correct parame-
Now we use the following expressions for extending ters) the RMSD should remain low (thin distributions).
the model to any Irradiance and Temperature conditions: Therefore do not confuse Model Accuracy and Parame-
ter Accuracy!

A 4.1 Incidence angle correction

1 = @ B C1,2/3 + D)< E4< 4<,2/3 FG
A2/3
(2) The irradiance on the module is determined by the
plane pyranometer measurement Glob . But we should
apply an Incident Angle correction (named IAM for
And for the diode saturation current: Incidence Angle Modifier) to the beam component,
accounting for the Fresnels laws about reflexions on the
H
4<  L L cover glass. We use the parameterization proposed by
I J "K P QR
? = ,2/3 @ B   !,MNO ! ASHRAE, i.e.
4<,2/3
(3)

^76 = 1 _? `1/ cos c 1)

(4)
where S is the Gap energy of the material (1.12 eV
for Si). with i = incidence angle, and the parameter _ =0.05.
We could not check this expression experimentally as
this information is too difficult to extract from the data
4. MODEL QUALITY CHARACTERIZATION
(the irradiance level effect mixes with this little effect).
The Global and Diffuse horizontal measurements are
Now we can apply the model for each (-./0( , 4./0( ) used for the determination of the beam component.
measurement, and compare the model results to the
measured /% characteristics. For the assessment of the
model, three distributions are analyzed: 5. RESULTS
- the maximum power U678 , which is of course the
basic result expected for use in the simulation of systems
with MPP tracking, 5.1 Methodology check with crystalline modules
- the short circuit current, )< , whose value is quasi- As a general calibration of our measurement pro-
identical to the photocurrent, cedure, we used several crystalline modules, for which
- the open circuit voltage, %V< , whose evolution is the standard model is reputed to be developed.
strongly determined by the internal behaviour of the We measured a Siemens M55 (Si-monocrystalline) dur-
model, especially according to temperature. ing one year, an Atlantis M55 (Si-mono) during 2.5
We avoid using the %6W and 6W distributions as final years, as well as a Kyocera (Si-poly) during 5 years (up
indicators, because they are not determined with high to now).
accuracy (they depend on the curve shape) and are Results of the pure Standard model for the Siemens
strongly correlated (only their product U678 is relevant). M55 are shown on Fig 1. We observe that the model
Observed distributions of differences between meas- underestimates the data at low irradiances.
ured and predicted data (often called errors) are ana- But when analysing the shunt resistance (which may
lysed as function of the relevant variables (irradiance, be measured on each I/V characteristics), we observe that
temperature) and quantified by the Mean Bias Differenc- it increases quasi-exponentially when the irradiance
es (MBD, noted ) and the standard deviation (the Root diminishes. If we model this behaviour (see next para-
Mean Square Differences RMSD, noted ). MBD and graph about amorphous), we obtain a flatter distribution,
and better difference indicators, for U678 as well as for
%V< errors. (see comment of the Fig. 1).
RMSD are usually expressed as percentage of the nomin-

The exponential ')* behaviour seems to be a general

al value (at STC).
These differences measurement minus model in-
clude not only the model inaccuracies, but also experi- rule: we observed it on all modules we have analysed.
This has an effect on the low-light irradiance perfor-
mance: higher ')* diminishes the associated loss, there-
mental uncertainties (irradiance and temperature mea-
surements, misalignment, variable albedo, shadings, dirt,
fore increasing the efficiency. This efficiency enhance-
ment is more pronounced when the ')* at STC is low, as
snow, etc).
The RMSD is representative of the spread of the re-
there are higher losses to be recovered.
The models parameters (') and ')* ) are completely
sults in different operating conditions. It is an indicator of
the validity of model when conditions are varying. The
MBE is sensitive to the primary parameters ()< , %V< , determined from the measured reference I/V characteris-
%6W , 6W ), i.e. on the chosen reference I/V characteristics. tics. But when dealing with manufacturers data the slope
NB: These indicators are referred as percentage of the around )< is not available, neither the full I/V curve, so
nominal values on each measurement. For getting an that we have to make hypothesis on these parameters.
error relatively to the energy yield, i.e. the average power
 25th European Photovoltaic Solar Energy Conference Valencia, Spain, 6-10 September 2010

Pmax Error, Meas - Model vs GlobP In a second step, we tried to find how the model pa-
6.0
rameters behave according to external conditions. Or
0 < Tmod < 80C Model
more generally, we looked for possible corrections to the
Pmax Error (Meas-Mod) [W]

4.0
model for matching our data. Three (sometimes four)
2.0
main modifications of the standard model are neces-
0.0 sary.
-2.0
1) Shunt resistance exponential behaviour
-4.0 As for crystalline modules the measured ')* is
strongly dependent on the irradiance level. The high-
irradiance ')* value is much lower with amorphous
-6.0
0 200 400 600 800 1000 1200
GlobP [W/m2]
technologies (the I/V slope around (5 is high), so that the
Figure 1: Errors distribution on U678 as function of -de_W for associated losses are very high. But the exponential
the Si-monocrystalline module M55. improvement toward low irradiances is also much more
Pure standard model: = 1.9 % and = 1.1 % on U678 pronounced (by a factor of 12 for the ')* `0)/')* (STC)
= 1.1 % and = 1.0 % on %V<
With ')* correction: = 0.2 % and = 1.2 % on U678
ratio, against 4 for crystalline modules).
The ')* distribution is shown on Fig. 3. We tried to
= 0.4 % and = 0.5 % on %V<
approximate it with a simple exponential expression:
5.2 Modelling a CIS module
We installed a CIS module (Shell ST40) since the ')* = ')* E-2/3 F + ')* `0) ')* E-2/3 F"
beginning of the project. Surprisingly, this module is hij S (5)
P K Q
quasi-perfectly described by the standard model, even  g SMNO ,
better than the crystalline one. CIS modules also need a where -2/3 is the irradiance for the reference I/V curve.
')* exponential correction, of the same order of magni-
tude as the crystalline modules. The results (measured-
R shunt function of Irradiance
model differences), shown on Fig. 2 for a 6-years period, 800
are impressive. They provide a valuable assessment of Measurements
the long-term stability of our experimental setup (the
R Shunt measured [ohm]

Parametrization
March-April 2005 deviation on fig. 2 is due to a dis- 600

placement of the temperature sensor).

400

Shell ST40 - CIS Seasonal effect

5.0% 200

4.0%

3.0%
Pmax (Meas - Model) difference

0
2.0% 0 200 400 600 800 1000 1200

1.0% Irradiance [W/m]

0.0% Figure 3: Amorphous triple-junction module SHR-17,

-1.0% ured ')* distribution and parameterization with an exponential
factor ')* = 5.5 and ')* `0) = 600 .
kl>
-2.0%

kl>
-3.0%

-4.0% We found that the value ')* = 5.5 actually pro-

-5.0%
vides a good approximation of the ')* data for most
janv 05

janv 06

janv 07

janv 08

janv 09

janv 10
sept 04

mai 05

sept 05

mai 06

sept 06

mai 07

sept 07

mai 08

sept 08

mai 09

sept 09

mai 10

sept 10

modules of different technologies we have tested. Hence,

the only parameter left in Eq. (5) is ')* `0). However this
kl>
is different for CdTe (')* 2.0) and micro-crystalline
Figure 2: Error distribution for ST40 (CIS) over 6 years.
= 0.2 % and = 1.0 % on U678
( 3.0).
Standard model
(with ')* correction): = 0.0 % and = 0.9 % on %V<
= 0.5 % and = 0.8 % on )< Using this corrected ')* when computing the model
in a simulation is the most effective correction to the
5.3 Amorphous triple junction standard model for representing amorphous modules.
The primary objective of this work was the modelling The measurement of the ')* value and its irradiance
of amorphous modules. Our first object of study was a behaviour are very easy, using any I/V measured curve.
triple-junction Unisolar tile module (SHR-17). With 3 These data are key parameters of the model, and should
superposed cells (sensitive in the blue for top, green- be part of the modules specifications in the future.
yellow for middle and red for bottom), these modules are 2) Recombination losses
not the simpler ones, but we had already some prelimi- While the standard model reproduces very well
the %?5 and %.> voltages for crystalline modules, it fails
nary results about them.
In a first step, we tried to apply the standard model
to predict the correct values for amorphous junctions. An
to such a complex system. A first observation was that
additional term in the general I/V equation was proposed
for any measured I/V characteristics, it is possible to find
by Merten et al. [3] in order to explicitly take the recom-
a set of parameters of the standard model, for which the
bination losses in the -i- layer of the p-i-n junction into
model perfectly matches the measured I/V curve. That
account. In this region, where takes place the main part of
means that the standard model is able to well represent
the photocurrent generation, the recombination of pairs is
the electrical behaviour. But, the problem is that a differ-
rather intense, fostered by the presence of dangling bonds
ent set of parameters is required for each (G, T) condi-
which act as recombination centres. This recombination
tions.
 25th European Photovoltaic Solar Energy Conference Valencia, Spain, 6-10 September 2010

current is in first approximation proportional to the When establishing the model for amorphous mod-
charge carrier concentration, and hence to the photocur- ules, we have now to determine 3 inter-dependent para-
rent. On the other hand, it is related to the electrical field meters (')* , '( and tv Dw), with a complex definition
in the c layer. This leads to the following expression for domain. We observed in all our modules that the tv Dw
the recombination current: parameter optimal value is high, around 80% to 90% of
its maximum value. But this maximum value is itself
strongly dependent on the ')* and ') choices.
2/5 = s tuv zCDw/33 E%xu `%  ') )FG
(6)
NB: The tv Dw parameter corresponding to our data
with does not fit the theoretical value proposed by Merten et
tuv : Thickness of the c layer (of the order of 0.3 D{),
al. [3]. Until now, we do not have any explanation for
Dw/33 : Effective diffusion length of the charge carrier,
that.
3) Spectral correction
D\ w\ D> w> In a-Si:H junctions, the gap energy ~S7W is around
Dw/33 = 2
D\ w\ + D> w> 1.6 eV, and therefore the spectral response of single
amorphous junctions is only sensitive to photons of high-
%xu : Intrinsic potential of the junction (Built-in vol- er energy, i.e. with wavelength  < 0.73 nm. Far red and
tage). Its value may be considered as constant, about IR photons are not energetic enough for creating an
0.9 % per junction (i.e. 2.7 % for triple junction). electron-hole pair.
The recombination current is a loss, which should be Because we could not perform spectral measurements
subtracted from the photocurrent. This corresponds to an in our experiment, we used a correction model proposed
additional term in the standard model. As it is voltage- by CREST1 of the University of Loughborough (Betts et
dependent it modifies the shape of the I/V curve (which al. [4]). This model is based on an estimation of the
does not match anymore exactly the I/V measurement), spectrum energy contents in the incident irradiance,
and therefore is not a perfect correction. But we keep it through the so-called APE variable (Average Photon
as it improves drastically the errors distribution, as it can Energy). Using one-year of spectral data, a Utilizing
be seen in Figs. 4a and 4b. Factor UF, fraction of the spectrum really effective, is
derived from the APE and the amorphous spectral re-
sponse function. This UF may be parameterized accord-
ing to known variables, i.e. the air mass and < (Clear-
Voc Model vs Voc measured
13

50 < GlobP < 1200 W/m Model ness Index normalised to Clear Day), as shown in Fig.
5.
12
Voc Model [V]

Utilization Factor a-Si:H 0.670-0.700

11
0.640-0.670
0.610-0.640
0.70 0.580-0.610
10
0.67 0.550-0.580

0.64 0.520-0.550
0.490-0.520
0.61
UF a-Si

9
0.460-0.490
9 10 11 12 13 0.58
Voc measured [V] 0.55
0.52
0.49
0.46
Voc Model vs Voc measured
13 0 0.2 1
2
0.4 3
0.6 4
50 < GlobP < 1200 W/m Model 0.8 5
KTc 1 6
Air Mass
12
Voc Model [V]

Figure 5: Utilisation Factor parameterization.

11

This correction is indeed not computed specifically

10
for the triple-junction spectral response. Nevertheless, we
decided to keep it as it slightly improves the final simula-
tion accuracy (improvement of 0.6% on and 0.4% on
9 ).
9 10 11 12 13
Voc measured [V] The final results on this triple-junction amorphous
Figures 4a and 4b: Distribution of the %?5 errors before and after
module are the following (full year 2009):
applying the recombination correction for the SHR-17 module.
= 0.1 % and = 2.3 % on U678
In our model, the quantity of recombination correc- = 0.3 % and = 0.5 % on %V<
tion is defined by the parameter tuv zCDw/33 G which we = 0.0 % and = 1.6 % on (5
consider as a new model parameter that we named Annealing seasonal effect
tv Dw . In our phenomenological approach, we establish This is of course not as good as than for crystalline
its value by minimizing the errors, especially on modules. But this technology is also subject to the Staeb-
the %V< distribution, but which also acts on U678 .
1
Centre for Renewable Energy Systems Technology
 25th European Photovoltaic Solar Energy Conference Valencia, Spain, 6-10 September 2010

ler-Wronski annealing seasonal variations, which is not

yet taken into account in our model. This effect is appar- 6. RESULTS ON ANY TECHNOLOGIES
ent on the 6-years evolution of the modules performance
on Fig. 6. It is higher, or of the same order of magnitude, Our test facility is now running since 6 years with 8
as the monthly accuracy of our model. channels available. We could experiment our model with
modules of all technologies available on the market.
Unisolar SHR-17 Wp Seasonal effect
10%
Crystalline modules: we calibrated our methodol-
8%
ogy with 2 monocrystalline (Siemens M55 and Atlantis
6% M55, and have now valuable data of a polycrystalline
Pmax (Meas - Model) difference

4% Kyocera module over 5 years, which does not show any

2% degradation.
0% CIS: In order to check a possible generalisation of
-2% the very good results on our old module, we have now a
-4% new CIGS module, ready to be installed on our test facili-
-6% ty.
-8% Amorphous, single junction: a little Flexcell
-10% module on flexible substrate, of VHF technologies.
sept 04
dc 04
mars 05
juin 05
sept 05
dc 05
mars 06
juin 06
sept 06
dc 06
mars 07
juin 07
sept 07
dc 07
mars 08
juin 08
sept 08
dc 08
mars 09
juin 09
sept 09
dc 09
mars 10
juin 10
6 years

Amorphous, tandem: Solarex MST-43MV and

Asiopack 30. A module EPV-40 was analysed as sample
Figure 6: SHR-17 results over 6 years, annealing effect.
Global results: = 0.7 % and = 2.8 % on U678
of one of our measured system.
Amorphous triple junction: besides the SHR-17,
= 0.7 % and = 0.7 % on Voc
we have data of a more recent model US-32 of Unisolar.
Micro-crystalline/amorphous: we have two mod-
4) Temperature correction (eventual)
The temperature behaviour on U678 , noted DU6WW
ules from Sharp in test since just one year, with only
preliminary results (due to initial degradation). We have a
[%/C] is normally a result of the model. But this does
module of another manufacturer to be measured soon.
not always match the specification of the manufacturer
CdTe: we have measured 2 modules of Firstsolar
(although it is probably a better estimation). Many people
since now 2 years. They behave in the same way as the
require that the model is perfectly in accordance with the
amorphous modules (with recombination correction and
specifications. Therefore we defined an additional correc-
annealing effect), but the spectral correction is not suited.
tion, i.e. a linear correction of the Gamma factor as func-
HIT (of Sanyo): we have a module ready to be
tion of the temperature, which affects the temperature
behaviour of the U678 and the %V< values. This correc-
measured, but not yet installed. Results should be availa-
tion allows to obtain any desired DU6WW value, but usual-
ble quickly as there is no initial degradation.
Fig. 7 summarizes the results of the differences (mea-
surement model) on U678 , %V< and )< from our long-
ly degrades the accuracy of the model (i.e. degrades the
value). Therefore we try to avoid using this artificial
term measurements of modules of any technologies.
correction when possible.

Figure 7. Long-term (Measurement-Model) results accuracy on modules of any technology (% of nominal values)
 25th European Photovoltaic Solar Energy Conference Valencia, Spain, 6-10 September 2010

7. CONCLUSIONS 8. REFERENCES

In this paper the accuracy of the standard one-diode [1] T.R. Betts, R. Gottschalg, D.G. Infield. Spectral Irradiance
Correction for PV system Yield Calculations, Proceedings of the
model for crystalline and CIS modules was assessed.
19th European Photovoltaic Solar Energy Conference, Paris,
With an exponential correction for the shunt resis- June 2004.
tance the accuracy of the RMSD between measured and [2] J. A. Duffie, W. -A. Beckman, Solar Engineering of Ther-
modelled power values () stays below 1.2% of the mal process, John Wiley and Sons, N-Y. 2\ Ed., 1991.
nominal power in any conditions over long periods (up to [3] J. Merten, J. M. Asensi, C. Voz, A.V. Shah, R. Platz, J.
6 years). Andreu, Improved Equivalent Circuit and Analytical Model for
Amorphous Silicon Solar Cells and Modules. IEEE Transac-
Trying to extend the standard model to amorphous
technologies, we found that besides the exponential ')*
tions on Electron Devices, Vol. 45, No 2, Feb. 1998
[4] A. Mermoud, Introduction des modules PV en couche
(which is the main correction), the standard model mince dans le logiciel PVsyst, final report of SIG-NER project,
requires two additional corrections: a recombination ISE, University of Geneva, May 2005,
loss term proposed by Merten et al. [3], and a spectral www.unige.ch/cuepe/html/biblio/pdf/THF_RapportFinal.pdf
correction computed by CREST (Betts et al. [1]). An [5] A. Mermoud, PVsyst: Software for the Study and Simula-
tion of Photovoltaic Systems, ISE, University of Geneva,
additional correction on the value may be used when
www.pvsyst.com.
necessary for temperature behaviour matching.
With our triple-junction module, these corrections
lead to an accuracy of = 2.3% over one year. But the
seasonal annealing effect which is not taken into ac-
count in our model dominates the effect of these correc-
tions. The monthly accuracies stay of the order of =
1.2%. All other amorphous modules lead to similar re-
sults.
The same model also applies to the CdTe technology
modules, but the spectral correction proposed is not
suited and is not used. The annealing effect is present but
less pronounced, so that the 18-months accuracy is 1.4%.
The results on our micro-crystalline modules are
slightly lower ( = 2.1% over 6 months), but are still
preliminary.
The corrections to the standard one-diode model pro-
posed in this paper are implemented in the PVsyst simu-
lation software (Mermoud [5]), developed by the Group
of Energy at the University of Geneva since 1994.