SAND2014-19059J

Modeling the Irradiance and Temperature Dependence of Photovoltaic

Modules in PVsyst
Kenneth J. Sauer1, Thomas Roessler2, Clifford W. Hansen3
1
Yingli Green Energy Americas, Inc., San Francisco, CA 94108, USA
2
Yingli Green Energy International AG, 8058 Zürich, Switzerland
3
Sandia National Laboratories, Albuquerque, NM 87185-1033, USA

Abstract — In order to reliably simulate the energy yield of

photovoltaic (PV) systems, it is necessary to have an accurate
model of how the PV modules perform with respect to irradiance
and cell temperature. Building on previous work that addresses
the irradiance dependence, two approaches to fit the temperature
dependence of module power in PVsyst have been developed and
are applied here to recent multi-irradiance and -temperature
data for a standard Yingli Solar PV module type. The results
demonstrate that it is possible to match the measured irradiance
and temperature dependence of PV modules in PVsyst.
Improvements in energy yield prediction using the optimized
models relative to the PVsyst standard model are considered
significant for decisions about project financing.
Index Terms — current-voltage characteristics, mathematical
model, minimization, photovoltaic cells, photovoltaic systems,
silicon, solar power generation.
Fig. 1. A 2D false color contour plot displaying the deviation Δηrel
of module efficiency relative to the value at STC for an example
I. INTRODUCTION crystalline silicon module type provides a visualization of the
dependence of this quantity on irradiance G and cell temperature T.
In order to make reliable statements regarding the expected
which is well established worldwide with independent
energy yield of photovoltaic (PV) systems based on available
engineering companies that specialize in the prediction and
meteorological data, it is necessary to know the dependence of
analysis of the energy yield of large PV systems.
the maximum power output Pmax of the modules comprising
A previous publication [4] describes a systematic approach
those systems on the two most important environmental
to determine four free empirical parameters of the one-diode
conditions: global plane-of-array irradiance G and operating
model used for modeling PV modules in PVsyst [5]. This
cell temperature T.
approach is based on an algorithm to minimize the root mean
To describe this dependence, it is convenient to use the
square deviation (RMSD) between modeled and measured
quantity Δηrel(G,T), i.e., the deviation of the module efficiency
module power as a function of irradiance only. In Ref. [4], it
at the given conditions G and T relative to the value at the
was assumed that the dependence of the module behavior on
corresponding reference conditions Gref and Tref, which is
temperature could then be reasonably well described by
calculated according to the formula
following the PVsyst standard approach of actively inserting
( )
( ) , (1) the nameplate value of the temperature coefficient of the
maximum power for the parameter µPmpp in PVsyst. For cases
where Pmax,ref is the maximum power output at the reference with no additional information about the temperature
conditions. Figure 1 visualizes this dependence for an example dependence available beyond the temperature coefficient of
module type, using a false color plot of Δηrel(G,T) with the Pmax, this is still a perfectly valid approach.
reference conditions here, and throughout the remainder of The main innovative aspect of the work presented here is
this paper, taken to be standard test conditions (STC), i.e., the fact that it overcomes this restriction, fully considering and
Gref = 1,000 W/m2, Tref = 25 °C, and the reference solar modeling the dependence of the module power on irradiance
spectrum [1]. and cell temperature. In fact, two different approaches to
Various commercial and non-commercial software model the temperature dependence more accurately are
programs are available for energy yield simulation, each with presented, including a simultaneous fitting of the dependence
a different application and purpose [2]. The work presented on irradiance and temperature. Furthermore, a detailed
here limits itself to the modeling and simulation of PV comparison of the quality of the model fits for different
modules of crystalline silicon technology in the commercial variants of modeling approaches and the corresponding impact
software program PVsyst from the University of Geneva [3], on energy yield estimates are given.
 Clearly, these model fitting tasks heavily depend on the the ratio of the reference irradiance Gref to the irradiance under
availability of suitable measurement data. In close cooperation test G can be replaced by the ratio of Isc values at the
with well known, independent PV testing institutes, methods corresponding irradiances to determine Δηrel(G,T), i.e.,
have been developed at the Yingli Americas PV Testing Lab ( ) ( ) ( )
(PVTL) in South San Francisco for collecting high quality ( ) . (2)
( )
measurements of the power output of PV modules across a
matrix of irradiance and cell temperature conditions with Once the values of Δηrel(G,T) have been determined for
acceptable effort and in good alignment with the testing each module, an overall average value is taken at each G and
practices outlined in the standard IEC 61853-1 [6]. This T condition, resulting in a single matrix of Δηrel(G,T) values
capacity fulfills the important need to collect such data representing the average performance of the sample. With
relatively easily for a statistically relevant sample of PV measurements at any given condition having a spread of at
modules, for rapid implementation into module performance most 1.1%, the average value is considered to be
models with reliable predictive power. representative of the entire sample.
In the following, it will first be explained how the This matrix of Δηrel(G,T) can be converted back to a matrix
measurement data recently collected at the PVTL and used for of Pmax(G,T) by rearranging Eq. (1) and solving for P max(G,T).
this study were obtained (Sec. II). Then, in Sec. III, it will be To obtain a matrix that is consistent with a specific nameplate
briefly reviewed how the dependence of the maximum module module power at the reference conditions, Pmax,ref is set equal
power on irradiance and temperature is realized in the one- to that power value. The resultant Pmax(G,T) matrix is used as
diode model in PVsyst. Section IV details how the correctness input for modeling and in all cases serves as the reference for
of the representation of the temperature dependence in PVsyst comparison with the modeled power across all G and T test
can be improved using straightforward means. Possible conditions.
modeling approaches are described in Sec. V, and the
corresponding fit results are compared and discussed in Sec. III. ONE-DIODE MODEL IN PVSYST
VI. In Sec. VII, the impact of the fit quality on energy yield
estimates is demonstrated for various geographical locations. The mathematical representation of the I-V curve in the
Finally, the work presented here will be summarized and an one-diode model is given in the most general form by the
outlook on future investigations will be sketched. following transcendental equation [4]-[5]:
[ ( )]
( ) { } , (3)
II. INPUT DATA FOR MODELING
where I and V are the current and voltage of the PV module,
The performance data used in this work were obtained respectively, Ncs is the number of cells in series in the PV
across a matrix of G and T conditions for a sample of twenty module, T is the cell temperature (K), q is the elementary
Yingli Solar 72 cell multicrystalline silicon PV modules of charge, and k is the Boltzmann constant. The free model
type YL310P-35b, with a nameplate power value of 310 W. parameters are thus initially the photoelectric current Iph, the
This sample was randomly selected from incoming shipments saturation current of the diode I0, the series resistance Rs, the
to the customers of Yingli Solar and delivered to the PVTL for shunt resistance Rsh, and the ideality factor of the diode γ.
indoor current-voltage characterization using a Class AAA In PVsyst, it is assumed that the photoelectric current is
flash tester with an integrated temperature control unit and dependent on G and T, as follows:
thermally insulated test chamber. Such an apparatus is
essential to ensure that a module under test reaches a ( ) [ ( )], (4)
homogeneous and steady state temperature condition prior to
electrical characterization [7]. where µIsc is the temperature coefficient of the short circuit
For each module, key points on the I-V curve, i.e., short current Isc of the module type under test. Also, the saturation
circuit current Isc, open circuit voltage Voc, and current Imp and current of the diode is assumed to vary with T as follows:
voltage Vmp at the maximum power point, were measured
[ ( )]
across a matrix of irradiance and cell temperature conditions ( ) ( ) , (5)
covering the ranges of 200 W/m2 to 1,000 W/m2 and 25 °C to
65 °C, in increments of 100 W/m2 and 5 °C, respectively. where εG is the energy of the band gap of the cell material.
Following IEC 61853-1, Section 8 [6], the so-called “self- Furthermore, the shunt resistance is assumed to vary with G,
reference method” can be applied for devices that are linear as follows:
with irradiance G in short circuit current Isc, in order to adjust
( )
for slight spectral mismatch or spatial non-uniformity effects ( ) ( ) , (6)
leading to less than 1% deviation from Isc linearity. Under
these conditions, which apply in the present case, in Eq. (1) with
 [( ) ], (7) Pmax(Tmin)
Pmax,ref
where Rsh,exp and the shunt resistance in the absence of Optimize µγ so that
irradiance Rsh,0 are two parameters describing this dependence Derivative µPmpp,slope ≈ µPmpp,NP
empirically [5]. equal to
In the original version of the PVsyst model, it is assumed µPmpp,NP
that the series resistance Rs and the diode ideality factor γ are
not dependent on G or T. Thus, this model requires values for Measurement Data
the following nine parameters: Iph,ref, µIsc, I0,ref, εG, Rs, Rsh,ref, Linear Fit to Measurement Data
Rsh,0, Rsh,exp, and γ. Two of these parameters, µIsc and εG, are Pmax(Tmax) PVsyst Standard Model (µ Pmpp,NP)
(µPmpp,NP)
assumed to be known a priori [5]. As described in Ref. [4], "Then T" Approach (Optimized µµγ)
γ)

provided values of Rs, Rsh,ref, Rsh,0, and Rsh,exp, the remaining Tmin Tref Tmax
model parameters can be computed. Consequently, there are
Fig. 2. Comparison of the dependence of the maximum power
four free parameters to fit the characterized behavior of a output Pmax on temperature T, with irradiance fixed at the reference
specific module. condition, as measured and as modeled in PVsyst. The squares
In Ref. [4], a systematic approach to determine these four represent the measurement data over the temperature range
parameters, with T fixed at Tref, is set forth in detail. Using [Tmin,Tmax], fitted linearly with the dash-dotted line to determine the
this approach, it is demonstrated that the measured slope corresponding to the measured or nameplate temperature
coefficient µPmpp,NP. For a PVsyst Standard Model, the additional free
dependence of the maximum power output of PV modules on empirical model parameter μ is adjusted such that the derivative of
G is faithfully reproduced and thus a considerable the modeled behavior, shown as the dashed line, with respect to T at
improvement of the fit quality over the PVsyst standard Tref equals µPmpp,NP. Over the entire temperature range, this clearly
modeling approach is achieved. The present work further does not provide as good a fit as the “Then T” approach, shown as
extends this approach by also considering the measured the solid line, which targets an agreement between the slope of the
model over this range and µPmpp,NP.
dependence of the maximum power output on T.
( ) ( ), (8)
IV. DEPENDENCE OF POWER ON TEMPERATURE IN PVSYST
replacing the model parameter γ with γref and introducing an
The dependence of the power output of a PV module on T is additional free parameter µγ, thus increasing the total number
usually assumed to be linear and is quantified by the relative of free parameters to five. In order to match the temperature
temperature coefficient of the maximum power expressed in dependence in the PVsyst standard modeling approach, μ in
units of %/°C. To determine this coefficient in accordance Eq. (8) is adjusted such that µPmpp,PVsyst is equal to the inserted
with IEC 60891 [8], measurements of the module power are measured or nameplate temperature coefficient value µPmpp,NP
taken over a temperature range of at least 30 °C, in steps of for the given module type [5]. However, because this
about 5 °C, with G fixed at Gref = 1,000 W/m2. Example procedure equates the derivative of Pmax(T) at T = Tref (i.e.,
measurements are indicated as squares in Fig. 2. The value of µPmpp,PVsyst) to a value derived from the slope of experimental
the temperature coefficient, denoted here as µPmpp,slope, is data over a range of temperatures (i.e., µPmpp,NP), as indicated
derived from the slope in W/°C of the linear fit over the full in Fig. 2, and since the derivative of Pmax(T) varies with T in
temperature range, shown as the dash-dotted line in Fig. 2, and the one-diode model [9], this results in discrepancies between
converted to relative terms of %/°C by dividing by Pmax,ref. the measured and modeled temperature dependence of the
For convenience, denote Pmax(Gref,T) simply as Pmax(T) in maximum power away from Tref. This fact therefore allows
the following. In PVsyst, the parameter called µPmpp, denoted some room for optimization of the free parameter µγ such that
herein for clarity as µPmpp,PVsyst, corresponds to the derivative the modeled Pmax(T) better matches the nearly linear measured
dPmax(T)/dT of the modeled Pmax(T) at T = Tref, rather than to behavior over the temperature range of interest.
the slope of the linear fit to the curve over some given The approach described in Ref. [4] can be extended
temperature range [9]. The basic one-diode model described in relatively easily in a modular way to optimize µγ accordingly.
equations (3)-(7) has the disadvantage that the derivative of To demonstrate this with a concrete example, the same
the modeled Pmax(T) at Tref is completely determined by the approach as described in Ref. [4], referred to here as the
choice of the free model parameters Rs, Rsh,ref, Rsh,0, and Rsh,exp “Only G” Model, is repeated to establish values for the free
and is often not consistent with the measured value [5]. model parameters Rs, Rsh,ref, Rsh,0, and Rsh,exp, with μ = 0.
Therefore, in PVsyst, in order to have more flexibility in Then, the value of μ is iteratively adjusted until the slope of
modeling the thermal behavior of the module power, a linear the modeled Pmax(T) agrees well with the slope of the
temperature dependence of the diode ideality factor γ was measured Pmax(T) over the full temperature range under test.
subsequently empirically accepted, as follows: The resulting modeled temperature dependence behaves
qualitatively like the solid line in Fig. 2 and, together with the
already fitted irradiance dependence, constitutes what is parameters Rs, Rsh,ref, Rsh,0, and Rsh,exp and is complemented by
referred to here as the “Only G - Then T” Model. inserting the measured temperature coefficient for µPmpp,PVsyst,
For the example data set described above, the measured from which μ is determined in the same manner as for the
values correspond to µPmpp,slope = 0.429 %/°C. In the PVsyst Standard Model.
remainder of this paper, this value is taken to be equal to
µPmpp,NP. However, when this value is inserted into PVsyst (as These three modeling approaches are explained above in
µPmpp,PVsyst) with the allowable precision of two decimals as of order of increasing number of optimized free parameters for
PVsyst Version 6.23 [3], one obtains a slope of 0.442 %/°C modeling the irradiance dependence of module power in
from the modeled Pmax(T) over the temperature range of the PVsyst, with zero for the PVsyst Standard Model, one (Rs) for
test data, overestimating the measured coefficient by nearly the PVsyst “V6 Tool” Model, and four (Rs, Rsh,ref, Rsh,0, and
3%. The explicit “Only G - Then T” optimization, on the other Rsh,exp) for the “Only G” Model. To each of these approaches,
hand, results in a coefficient of 0.430 %/°C, almost an exact the “Then T” approach to determine the additional free model
reproduction of the measured value. parameter μ described in Sec. IV can be added, in order to
Therefore, by adding this “Then T” approach as an better match the measured temperature dependence, resulting
extension to the systematic methodology described in Ref. [4], in three additional modeling approaches.
the measured dependence of the maximum power output of a Clearly, with all of these modeling approaches, no more
module on cell temperature T can also be reproduced very than one row and one column of the matrix of measured
well in a PVsyst model. values of Pmax(G,T) are taken into account when fitting the
free model parameters. The row and column correspond to the
measured power values at Tref at all available irradiances, and
V. OVERVIEW OF MODELING APPROACHES
the measured power values at Gref at all available cell
The “Then T” approach to determine the value of the free temperatures. It is obviously questionable whether the PVsyst
module parameter μ to fit the measured temperature models that result from these modeling approaches achieve the
coefficient described in Sec. IV in the context of the so-called best global agreement of modeled and measured values of
“Only G - Then T” Model can be combined with any approach Pmax(G,T) across the entire G-T variable space of conditions
to modeling the module behavior at 25 °C that results in under test, given that these approaches use only a subset of the
values for the free model parameters Rs, Rsh,ref, Rsh,0, and available measurement data. To explore this, a numerically
Rsh,exp. In the following, several of the most frequently used expensive simultaneous adjustment of all five free model
modeling approaches are described and the extensions made parameters, including μ, to the full matrix of measured
possible by this work are explained. Pmax(G,T) values, hereinafter referred to as the “G-T Matrix”
Model, is required. A numerical approach to the parameter
A. PVsyst Standard Model
optimization, involving the trust-region-reflective algorithm
In order to generate what is denoted here the “PVsyst related to the Levenberg-Marquardt method for
Standard Model” for the module type under test, default multidimensional minimization, was implemented to make the
values prescribed by PVsyst are used for the parameters Rs, Parameters for
Rsh,ref, Rsh,0, and Rsh,exp. As mentioned earlier, in this model the Modeling T

measured, or, equivalently here, the nameplate temperature

coefficient of the maximum power is also inserted into PVsyst
as µPmpp,PVsyst in order to determine the value for μ by setting Fit m
dPmax(T)/dT = µPmpp,PVsyst at T = Tref, which in turn determines
the temperature dependence of the modeled Pmax(T). Nameplate /
Measured
B. PVsyst “V6 Tool” Model µPmpp
Parameters for
As of Version 6, PVsyst itself offers a tool to adjust the free Default Fit Rs Fit Modeling G

model parameter Rs to fit to a measured irradiance dependence Rs, Rsh,ref, plus Default Rs, Rsh,ref,
Rsh,0, Rsh,exp Rsh,ref, Rsh,0, Rsh,exp Rsh,0, Rsh,exp
of the module power, while keeping the parameters Rsh,ref,
Fig. 3. This plot provides an illustration of the various modeling
Rsh,0, and Rsh,exp at the default values for the given module approaches explored in this work, sorted by the number of fitted free
type. This approach, combined with setting µPmpp,PVsyst to the parameters for modeling the irradiance G (x-axis) and temperature T
measured temperature coefficient as for the PVsyst Standard (y-axis) behavior of PV modules in the one-diode model in PVsyst.
Model, constitutes the PVsyst “V6 Tool” Model. The “Then T” approach introduced in this work to adjust µγ to better
match the measured temperature dependence of PV modules (red)
C. “Only G” Model can be added as a modular extension to any of the three approaches
to modeling the multi-irradiance behavior at 25 °C (blue), resulting in
The “Only G” Model corresponds to the modeling approach six modeling approaches plus the simultaneous optimization of the G
described in Ref. [4], which determines the free model and T dependence, referred to here as the “G-T Matrix” Model.
corresponding determination of a global minimum in the 5D Measurement Data PVsyst Standard Model
model parameter space feasible. PVsyst "V6 Tool" Model "Only G - Then T" Model
The hierarchy of the seven modeling approaches described "G-T Matrix" Model
above is visualized in Fig. 3, explicitly using the number of 0%
optimized free parameters for modeling the G and T
-5%
dependence as a way to distinguish between them.
-10%

Δηrel
VI. FIT RESULTS -15%

All seven modeling approaches discussed in Sec. V were -20%

applied to obtain models for the example data set described in
-25%
Sec. II. Table I compares the RMSD and mean bias error

1000

1000
200

400

600

800

300

500

700

900

200

400

600

800
(MBE) between the modeled and measured maximum power
across the entire G-T variable space, with the convention for 25 45 65
calculating residuals here, and throughout the remainder of G [W/m2] // T [ C]
this paper, of modeled minus measured power. Also compared Fig. 4. Plot of Δηrel(G,T) curves at a selection of cell temperatures
in Table I are relations to key nameplate quantities, i.e., the including 25 °C, 45 °C, and 65 °C for the measurements discussed
percent deviation from µPmpp,NP and the difference from the here, as well as the PVsyst Standard Model for the module type
average measured Δηrel at the low irradiance condition (LIC) under test and the PVsyst “V6 Tool”, “Only G - Then T”, and “G-T
Matrix” Models with optimized settings. Clearly, the more advanced
of 200 W/m2 irradiance, 25 °C cell temperature, and the models, especially the latter two introduced in this work, are more
reference solar spectrum [6], denoted here as Δηrel,LIC, which is reflective of the measurement data compared to the PVsyst Standard
reported on datasheets in accordance with EN 50380 [10] and Model, as can be seen with some distinction at higher temperatures.
for this sample almost exactly equals the nameplate value of
global fit is obtained, as expected, by the “G-T Matrix”
3.3%. For the PVsyst “V6 Tool” Model, the RMSD value of
Model, with an RMSD of 0.25 W and an MBE of 0.18 W.
0.97 W is almost a factor of 2 smaller than the RMSD of
From Table I, where the modeling approaches are listed in
1.92 W resulting from the PVsyst Standard Model. While this
order of increasing RMSD, it can be seen that the global
is a clear improvement of the global fit quality, the negative
model fit quality increases with the number of optimized free
MBE value of 0.79 W for the PVsyst “V6 Tool” Model still
model parameters, with the “Then T” approach to fitting the
indicates a systematic underestimation of the module power
temperature dependence providing significant improvement,
(as is also apparent from Fig. 5, discussed below). The
and the “G-T Matrix” Model still with an additional edge. On
“Only G - Then T” Model, with an RMSD of 0.40 W and an
the other hand, in terms of reproducing more closely the
MBE of 0.33 W, improves the fit quality by another factor of
nameplate parameters µPmpp,NP and Δηrel,LIC, the “Only G -
more than 2 over the PVsyst “V6 Tool” Model, but the best
Then T” Model yields the best results, with a relative
TABLE I deviation of 0.04% and a difference of 0.01%, respectively,
COMPARISON OF MODELING APPROACHES compared to 0.78% and 0.16% for the “G-T Matrix” Model.
It is important to note that the question about the “best”
Modeling RMSD MBE Dev. from Diff. from
modeling approach cannot be answered unequivocally, as the
Approach [W] [W] µPmpp,NP [%] Δηrel,LIC [%]
choice depends on the planned usage of the model.
G-T Matrix 0.25 0.18 0.78 0.16 For a visualization of the most important results, Fig. 4
compares modeled Δηrel(G,T) at all irradiances and at a
selection of cell temperatures to the average measurement data
Only G Then T 0.40 0.33 0.04 0.01 (squares). The PVsyst “V6 Tool” Model (plus signs) and, even
more so, the “Only G - Then T” (triangles) and “G-T Matrix”
V6 Tool Then T 0.47 0.39 0.52 0.12 (crosses) Models introduced in this paper, clearly describe the
measured behavior more faithfully than does the PVsyst
Input Standard Model (diamonds). From the graph, the remaining
Only G 0.89 0.73 2.68 0.01
µPmpp,NP differences between the models resulting from these three
Input more advanced modeling approaches can at most be seen
V6 Tool 0.97 0.79 3.19 0.12
µPmpp,NP qualitatively, in particular at the higher temperatures.
Standard Figure 5 provides further insight into the differences
Then T 1.52 1.39 0.27 1.59 between modeling approaches, using a false color contour plot
Model
of power residuals for the same four models. The PVsyst
Standard Input Standard Model exhibits large negative residuals in particular
1.92 1.79 2.96 1.59
Model µPmpp,NP at intermediate irradiances and higher temperatures. In
 PVsyst “V6 Tool”, “Only G - Then T”, and “G-T Matrix”
Models, with all other modeling parameters held fixed or at
the PVsyst default settings.
The results shown in Fig. 6 are reported as deviations of the
annual forecasted energy yields (in this case, gains) from those
simulated using the PVsyst Standard Model, using PVsyst
default settings for losses. Because the impact of a specific
modeling approach on the estimated module and, thus, system
performance at a given irradiance and temperature scales with
the frequency of occurrence of those conditions in the
simulated meteorological year, some variance is expected in
the results for a given model across the different geographical
locations.
It is apparent that the models resulting from the more
advanced approaches predict approximately 0.8-1.0% more
Fig. 5. False color contour plot of maximum power Pmax residuals, annual energy yield than the PVsyst Standard Model.
with the convention of modeled minus measured power, as a function Furthermore, for the three lower latitude locations, the models
of irradiance G and cell temperature T across the entire G-T variable resulting from the two approaches introduced in this work
space of measurement conditions for the nameplate 310 W Yingli
Solar module type under test. While the PVsyst “V6 Tool” Model systematically predict a 0.20-0.25% higher annual energy
shows an improvement over the PVsyst Standard Model, resulting in yield than the PVsyst “V6 Tool” Model. This is consistent
smaller power residuals, clearly the distributions of residuals for the with the fact that the PVsyst “V6 Tool” Model underestimates
“Only G - Then T” and “G-T Matrix” Models arising from this work the module power at the higher temperatures predominant at
are more randomly distributed and uniformly close to zero, indicating these locations, as seen in the upper right false color plot of
a globally well-fitting model.
power residuals in Fig. 5. Although these differences may
comparison, the PVsyst “V6 Tool” Model is already greatly seem minor compared to other uncertainties in modeling the
improved, but the residuals increase systematically for cell performance of a PV system, for project financing, differences
temperature conditions further away from T = 25 °C at all even of this magnitude can be decisive.
irradiances. This is expected, since only multi-irradiance data For Berlin, Germany, where conditions of low temperature
at that temperature were used to adjust model parameters. and irradiance make a more significant contribution to the
Clearly, the residuals for the “Only G - Then T” Model and, annual energy yield, the results are not as easily explained.
even more so, for the “G-T Matrix” Model are more randomly Possibly, the overestimation of the module power for cell
distributed and uniformly close to zero, indicating a better temperatures below 25 °C results in a relative increase in the
global fit to the data. 1.6%
It is noteworthy that the values for the difference from the PVsyst "V6 Tool" Model
average measured Δηrel,LIC in the last column of Table I clearly 1.4%
Gain in Forecasted Annual

"Only G - Then T" Model

indicate that the PVsyst “V6 Tool” Model considerably 1.2%
improves the fit to the irradiance behavior at 200 W/m2 as "G-T Matrix" Model
Energy Yield

1.0%
compared to the PVsyst Standard Model scrutinized in Ref.
[4]. Looking at the Pmax residuals along the x-axis of the false 0.8%
color plots in Fig. 5, it is obvious that this still holds true,
0.6%
albeit to a lesser extent for intermediate irradiances. However,
only the “Only G - Then T” and “G-T Matrix” Models 0.4%
introduced in this work achieve a good fit over the entire 0.2%
irradiance range considered.
0.0%
Antofagasta, Berlin, Phoenix, Trivandrum,
VII. IMPACT ON ENERGY YIELD ESTIMATES Chile Germany USA India
Fig. 6. Four geographically distributed locations were chosen as
PVsyst was used to generate annual energy yield forecasts examples for assessing the deviations of the forecasted energy yields
for a fixed PV system design at a diversity of geographical obtained with the PVsyst “V6 Tool”, “Only G - Then T”, and “G-T
locations covering a wide range of climatic conditions, for the Matrix” Models discussed in this work, relative to the PVsyst
Yingli Solar module type under test. Baseline simulations Standard Model. The impact of a specific modeling approach on the
estimated module and, thus, system output at a given irradiance and
using the PVsyst Standard Model were run at each location at temperature will scale with the frequency of those conditions in the
the optimal tilt angle reported by PVsyst and compared to simulated meteorological year. Some variance in the results reported
simulations using the parameter values determined for the for a given model across the different locations is therefore expected.
energy yield as simulated with the PVsyst “V6 Tool” Model, in noteworthy gains of simulated annual energy yield on the
whereas the underestimation at conditions near LIC reduces order of 0.8-1.0% at diverse geographical locations.
the prediction in the case of the “G-T Matrix” Model. For this Additional studies with measurement data from other
location, it may be necessary to extend the range of measured module types and technologies are required to test the general
temperature conditions to values below 25 °C and perform a applicability of these methodologies. This will also prove
more detailed analysis of the contribution of the low whether consistently better global fits can be obtained by the
temperature and irradiance conditions to gain a better “G-T Matrix” Model compared to the “Only G - Then T”
understanding of the results. Model, which may justify the additional measurement and
computational expense of the former.
The capability to generate a PVsyst model based on the
VIII. SUMMARY AND OUTLOOK
“G-T Matrix” approach now opens several enticing avenues of
For reliable energy yield simulation, it is necessary to have investigation. It would, for instance, be interesting to combine
an accurate module model of Pmax(G,T). Reference [4] this approach with a weighting of environmental conditions
presents a method for the systematic adjustment of parameters according to their contribution to the annual energy yield at a
of the one-diode model in PVsyst to reproduce the measured specific location to build what could be denoted a “location-
dependence of the maximum power output of a PV module on optimized” PVsyst model. Additionally, it would be
the irradiance G. Building on this, two approaches have been interesting to determine the minimum conditions for module
developed to better represent the dependence of the maximum measurements that are required to obtain a PVsyst model with
power output on cell temperature T. These were applied to acceptable accuracy over the full ranges of irradiance and
model high quality module power measurements for a sample temperature.
of Yingli Solar multicrystalline PV modules.
The first such approach, denoted as “Then T”, can be
ACKNOWLEDGEMENTS
combined with any approach to modeling the module power at
25 °C as a function of irradiance that results in values for the The authors thank Junaid H. Fatehi for always enlightening
free model parameters Rs, Rsh,ref, Rsh,0, and Rsh,exp. The discussions.
“Then T” approach obtains an excellent model fit of the
temperature coefficient of the maximum power, derived from
REFERENCES
the measured slope of Pmax(T) over the full temperature range
under test. Combined with the “Only G” approach to fitting [1] Standard Tables for Reference Solar Spectral Irradiances:
the irradiance dependence of the module power introduced in Direct Normal and Hemispherical on 37° Tilted Surface, ASTM
G173-03, 2012.
Ref. [4], this reproduces the global dependence of the
maximum power output on both irradiance and temperature. [2] G. T. Klise and J. S. Stein, “Models used to assess the
As of Version 6.26, released shortly after the submission of performance of photovoltaic systems,” Sandia National Lab.,
this paper, the modeling of the temperature dependence in Albuquerque, NM, Rep. SAND2009-8258, 2009.
PVsyst has reportedly been improved in line with the
[3] A. Mermoud. (1994-2014). PVsyst (Version 6.23). [Computer
proposed “Then T” approach presented here. Software]. Geneva, Switzerland: ISE, University of Geneva.
The second approach, termed the “G-T Matrix” approach, Retrieved May 1, 2014. Available from www.pvsyst.com.
determines all five free model parameters simultaneously by
optimizing the fit of the model to the measured maximum [4] K. J. Sauer and T. Roessler, “Systematic approaches to ensure
power output across the full matrix of irradiance and correct representation of measured multi-irradiance module
performance in PV system energy production forecasting
temperature conditions. For the example data set used here, software programs,” IEEE J. Photovoltaics, vol. 3, pp. 422-428,
the “G-T Matrix” approach improved the global fit of the Jan. 2013.
model, as expected, but at the expense of accurately modeling
two key nameplate quantities, i.e., the temperature coefficient [5] A. Mermoud and T. Lejeune, “Performance assessment of a
of the maximum power and the deviation of the module simulation model for PV modules of any available technology,”
in 25th European Photovoltaic Solar Energy Conference,
efficiency at the low irradiance condition, i.e., 200 W/m2, Valencia (Spain), pp. 4786-4791, 2010.
25 °C, and the reference solar spectrum, relative to the value
at STC. [6] Photovoltaic (PV) module performance testing and energy
Either of these approaches significantly improves the global rating – Part 1: Irradiance and temperature performance
fit quality of the resulting models compared to the PVsyst measurements and power rating, IEC 61853-1 Ed. 1.0, 2011.
Standard Model and its variants. Thus in principle this work [7] M. Joshi and R. Singh, “Toward reliable module temperature
has demonstrated that it is possible to reliably reproduce in measurement: Considerations for indoor performance testing,”
PVsyst the measured irradiance and temperature behavior of a presented at the Sandia/EPRI 2014 PV Systems Symposium: 3rd
PV module. These model improvements were shown to result PV Performance Modeling Workshop, Santa Clara, CA, 2014.
[8] Photovoltaic devices – Procedures for temperature and
irradiance corrections to measured I-V characteristics, IEC
60891 Ed. 2.0, 2009.

[9] A. Mermoud, private communication, July, 2011.

[10] Datasheet and nameplate information for photovoltaic modules,

EN 50380, 2003.