Anda di halaman 1dari 8

25th European Photovoltaic Solar Energy Conference and Exhibition /

5th World Conference on Photovoltaic Energy Conversion, 6-10 September 2010, Valencia, Spain

TWO-DIODE MODEL REVISITED:


PARAMETERS EXTRACTION FROM SEMI-LOG PLOTS OF I-V DATA

G.H. Yordanov1,2, O.M. Midtgård1 and T.O. Saetre1


1
University of Agder, Jon Lilletuns vei 9, 4879 Grimstad, Norway, Tel: +47 37233440, Fax: +47 37233001, E-mail:
georgihy@uia.no 2NTNU, O.S. Bragstadsplass 2E, 7491 Trondheim, Norway

ABSTRACT: This paper presents an improved method to extract physically meaningful parameters from outdoor I-V
data of polycrystalline-Si PV modules. The two-diode model of a PV cell is adapted for modules. A novel algorithm is
devised to estimate the series resistance from I-V data sets. The ideality factor and the reverse saturation current of the
diffusion diodes are then calculated from partially linear semi-logarithmic plots. The complexity and ambiguity typical
for non-linear curve fitting are thus avoided. The method is applied to outdoor I-V curves from polycrystalline-Si
modules, and preliminary results are presented. A new definition of the local ideality factor is introduced. It compensates
for the series-resistance effect at higher voltages, such that a constant ideality factor is obtained for polycrystalline-Si
modules in this range of voltages.
Keywords: Module Series Resistance, p-n Junction Ideality Factor, Polycrystalline Silicon

1 INTRODUCTION contributions of the present work.

The present paper applies a novel technique, which


we first introduced in [1], for determination of the series 2 REVIEW OF THE TWO-DIODE MODEL
resistance of crystalline-Si PV modules from I-V curves.
The series resistance is then used to further estimate 2.1 Applications for crystalline Si dominate literature
other parameters from the two-diode model (also called The two-diode model has primarily been applied to
‘the double-exponential model’) of a PV cell, here monocrystalline-Si PV cells [2-8], as this technology
adapted for PV modules. Linear fitting of semi- dominated the market for the initial half century of the
logarithmic plots is applied. This approach avoids the modern photovoltaic age [9,10]. There are few reports on
ambiguity and complexity (mathematical as well as the application of the model for polycrystalline cells.
computational) of the non-linear curve fitting which is However, due to its lower production cost,
widely used in the available literature. polycrystalline-Si has recently gained market shares
The method presented here is applied to outdoor I-V similar to monocrystalline-Si. Therefore, it is important
data of similar modules with screen-printed to study the applicability of the two-diode module for
polycrystalline-Si cells, based on two types of Si- polycrystalline-Si cells as well.
feedstock. The preliminary results for the reverse
saturation current I01 correspond well with theory. 2.2 The lumped circuit model
We also introduce a new definition for the local The two-diode, lumped-circuit model of a PV cell is
ideality factor with compensation for the series-resistance shown in Fig. 1. The mathematical form of the model is
effect. We show that for polycrystalline-Si modules the given by Eq. (1).
modified m-V plots will then contain a region of constant
ideality factor at the higher voltages.
RS I
Furthermore, to our knowledge, this is the first IPH D1 D2 +
application of the double-exponential model to a PV
module based on solar-grade silicon (SoG-Si) produced RP V
from a metallurgical route. The comparison with a
similar module based on SoG-Si purified by the classical _
chemical route, shows no practical difference in terms of
the parameters obtained.
Figure 1: Lumped-circuit, two-diode model of a PV cell.
1.1 Organisation of the paper
The paper begins with a review of the two-diode I (V ) = IPH − IP − ID1 − ID2 =
model and its mathematical formulations. Possible
approaches for extracting the unknown parameters are V + IRS  VnV
+IRS
  Vn+VIRS  (1)
then discussed. Thereafter, we examine semi-logarithmic IPH − − I01  e 1 T
−1 − I02  e 2 T −1
RP    
plots of experimental I-V data, and observe their    
behaviour at the higher end of voltages. A new algorithm
for the estimation of the series resistance is then The terminal current I of the cell can be divided into
explained in detail. four components, as shown in Eq. (1): the
We then present and discuss the results obtained for photogenerated current (IPH), the current through the
the parameters. Special attention is given to the values shunt resistance (IP), the diffusion-diode current (ID1),
obtained for the diffusion-diode ideality factor. We and the recombination-diode current (ID2). The 7
present a new definition of the local ideality factor where unknown parameters in the model are: the
the m-V plots are compensated for series-resistance photogenerated current IPH; the series resistance RS; the
effects. These plots provide a good approximation to the shunt (or parallel) resistance RP; the reverse saturation
ideality factor n1. Finally, we summarize the

4156
25th European Photovoltaic Solar Energy Conference and Exhibition /
5th World Conference on Photovoltaic Energy Conversion, 6-10 September 2010, Valencia, Spain

current I01 and the ideality factor n1 of the diffusion parameters for cells and modules less straight-forward.
diode; and the reverse saturation current I02 and the Lo Brano et al. (2010) have used an ideality factor of
ideality factor n2 of the recombination diode. n1 is often dimension Volts/Kelvin incorporating the elementary
assumed to be equal to 1 by many authors, in accordance charge and Boltzmann’s constant [14]. By stripping qe
with the diffusion theory of p-n junctions [11], whereas and kB from their ideality-factor estimates, we have
n2 is sometimes set equal to 2, in accordance with the arrived at classical (dimensionless) ideality factors of
theory of recombination via traps [12]. The thermal 0.87 and 0.73, which are not physical.
voltage VT=kBT/qe, where T is the p-n junction 3D effects in solar cells have been investigated as
temperature (considered to be a known or controlled early as in 1963 [4]. When evaluating distributed-series-
quantity), kB is Boltzmann’s constant, and qe is the resistance effects or current-flow patterns (e.g. current
elementary charge. crowding), the two-diode model shown in Fig. 1 becomes
The parameters in the two-diode model depend on too simple to accurately model the output from a PV cell.
the irradiance and cell temperature [6,7,8]. At least one more diode and one more resistor must be
added when studying localized loss mechanisms, e.g.
2.3 Adaptation to modules resistance-limited enhanced recombination [15]. For the
A few papers on parameter-extraction from module I- PV modules tested here, we assume that none of their
V data have been published, see e.g. [5,13]. Such studies cells requires such complicated modelling when
must consider that there are many tens of individual cells illuminated outdoors. Experimental results described in
in an average size module, and the cells may not be Sections 5 and 6 confirm this assumption.
absolutely identical in terms of their parameters. Modules
also incorporate additional semiconductor elements
(bypass diodes) that further complicate the modelling of 3 PARAMETER-EXTRACTION APPROACHES
their electrical characteristics. However, with cells that
are nearly identical, the mathematical form of the two- 3.1 Non-linear fitting of I-V curves
diode model can be adapted for PV modules. One can RP of a cell can be found from a linear fitting of an I-
then extract module-averaged parameters, assuming V curve near short circuit. For modules this approach
minor variance of their values among the individual cells. does not always work, as bypass diodes may be activated
The results from this work indicate that for the industrial at lower voltages, introducing specific kinks (see Fig. 2).
PV modules tested, this assumption is valid. The approximation IPH≈ISC is generally valid for most
Most modern polycrystalline-Si modules consist of crystalline-Si cells and modules, as I01 and I02 are several
large-area cells connected in series, without paralleling. orders of magnitude smaller than IPH, and also RS<<RP.
This is also true for the PV modules tested in this work. However, the estimation of the remaining 5 parameters (3
One can adapt Eq. (1) to the form given by Eq. (2) of which is the focus of this paper) is not similarly
assuming that all cells are nearly identical. V and I are straight-forward. Most authors seem to prefer non-linear
now the terminal voltage and current of the module, curve-fitting approaches to estimate all parameters [2-8].
respectively. NS is the number of cells in series. Some assume a priori integer values n1=1 and often also
n2=2, and evaluate the other parameters by fitting. Initial
I (V ) = guesses (starting values) are required by most of the non-
linear fitting approaches.
V + IRS  VN+nV
IRS
  NV +nIRVS  (2) After reviewing many earlier works on non-linear
IPH − − I01  e S 1 T
−1 − I02  e S 2 T −1 fitting, Appelbaum et al. (1993) concluded that there can
RP     be wide parameter sets all resulting in satisfactory curve
    fitting [2]. A question then arises about the physical
significance of the parameters obtained using one or
2.4 Model variants found in the literature another technique, since the two-diode model
Watson (1960) first proposed a two-diode model for approximates the physical mechanisms of conduction in
crystalline-Si PV cells, but with identical exponents [4]. crystalline-Si PV cells. Besides, some of the parameters,
Wolf and Rauschenbach (1963) focused on distributed- such as RS, RP and I02, may vary depending on the
series-resistance effects in PV cells, and arrived at an operating point on the I-V curve. If care is not taken to
equation similar to Eq. (1) [4], but they ignored the shunt find a representative value, incorrect values for the other
current due to the high shunt resistance (RP=13.8 kΩ) in parameters will be obtained [6,16,17]. The non-linear
their experimental 2-cm2 monocrystalline-Si cell. curve-fitting procedures are also quite complicated both
Müllejans et al. (2004) preferred a 5-parameter mathematically and in terms of computer code, a
model, assuming a priori n1=1 and n2=2 [5]. They complexity that should be avoided if possible. Gottschalg
reported RP-values of practical (large-area) poly-Si PV et al. (1997) calculated the linear parameters (IPH, 1/RP,
cells as low as 6.6 Ω. Other authors have also applied 5- I01 and I02) and fitted only the non-linear ones (RS, n1 and
parameter versions of the model [6,8,13]. Some have n2) [3]. However, this approach is difficult to program
considered the recombination diode negligible. Coors and computationally intensive.
and Böhm (1997) have found the more complete 7-
parameter variant superior to IEC-891 and Blaesser’s 3.2 Partial linear fitting of semi-log I-V plots
procedure in I-V curves correction [7]. Ouennoughi and Some authors have used semi-logarithmic plots of I-
Chegaar (1999) have used the single-diode V data to analyze p-n junctions [4,13,18,19]. It has been
approximation [13]. They have applied the same equation possible to distinguish linear portions of semi-
to a PV module, leaving the ideality factor to also logarithmic plots where one of the two exponential terms
account for the number of cells in series. Such an dominates. The slope of such a linear region equals
approach makes the direct comparison of p-n-junction qe/nkVT (qe/NSnkVT for modules) which is used to extract

4157
25th European Photovoltaic Solar Energy Conference and Exhibition /
5th World Conference on Photovoltaic Energy Conversion, 6-10 September 2010, Valencia, Spain

nk, k=1,2; whereas the current-axis intercept of the linear chemical route. Cells and modules were manufactured by
fit gives the reverse saturation current I0k, k=1,2. Wolf Q-Cells. The modules are installed on a flat roof at
and Rauschenbach (1963) have found that the latitude N58°09’ in the city of Kristiansand, Norway,
recombination-diode ideality factor of their cell is close facing South. The inclination angle is 60° from the
to 3 and temperature-dependent [4]. It is known that, if horizontal (optimized for spring and autumn generation).
the plots are not correctly compensated for series- Fig. 2 shows non-shaded outdoor I-V curves of the
resistance effects, characteristic deviations from linearity modules tested. Activation of bypass diodes at high
are observed at higher voltages [4,18]. The next Sections currents is apparent in the I-V curve of the module
describe our novel algorithm to accurately estimate RS denoted A10162 due to slight current mismatch of its
from an I-V data set, and how we have adapted the cells. However, this takes place at a region of the I-V
methodology of RS-compensated semi-logarithmic plots curve that is not so relevant for our analysis, as will be
to extract n1 and I01. seen in Section 5.1.
McIntosh (2001) concluded that physically consistent
parameters could not be extracted from tangents to semi-
logarithmic I-V plots, as there might be several
conduction and recombination mechanisms present, none
of which dominate any part of the curve [15]. That study,
however, focused on buried-contact cells which featured
additional detrimental effects due to e.g. ill-passivated
edges, floating junctions and/or localized Schottky
contacts. Analysis of the local ideality factor m (m-V
plots) was the key method applied by McIntosh.
Although the cells were modelled with the two-diode
model assuming n1=1, the m-V plots (modelled as well as
experimental) did not have regions where m=1. One
reason for this is that, by definition, the m-V plots are not
compensated for the series-resistance effect (as RS is
generally not known in advance). Another reason is that
the detrimental effects studied took place at moderate Figure 2: Outdoor I-V curves of the 3 polycrystalline-Si
voltages, thus interfering with the series-resistance effect. PV modules tested.
In general, all detrimental mechanisms (be it shunting,
resistive losses or some type of recombination) lead to an 4.2 Data-acquisition system
increase in the local ideality factor to well above m=1. We used the commercial multichannel electronic load
In Section 6.2.1 of this paper we apply m-V plots system Dynaload MCL488 controlled by custom-made
with added compensation for the series-resistance effect. Labview code via the data-acquisition board NI PCIe-
At higher voltages these improved m-V plots contain a 6363 (PC-based). The system only permits backward I-V
region of practically constant local ideality factor, sweeps (starting from open circuit). The on-state
slightly higher than 1. resistances of the electronic loads are about 0.15 Ω.
Complete shorting of the modules is therefore not
3.3 Dark and multiple-quadrant I-V analysis possible, as seen in Fig. 2. Extrapolation is therefore used
The two approaches reviewed above are also to estimate ISC.
applicable to dark I-V curves, sometimes used in The in-plane irradiance during sweeps was measured
combination with illuminated characteristics to find by a polycrystalline-Si reference cell SolData 80spc.
particular parameters, RS being the most relevant However, the three modules tested have been subject to
example. An inherent problem with this approach is the reference measurements by an independent, certified lab,
inaccurate determination of RS due to the different and since they all gave self-referenced irradiance-values
current-flow patterns (e.g. current crowding) in dark from ISC that were practically equal (within 0.1% at 1000
versus illuminated cells [20]. Reverse I-V characteristics W/m2), we considered these self-referenced values to be
are also used to find individual parameters, see e.g. [4]. more reliable, and used them instead.
Dark and multiple-quadrant methods are, however, I-V curves of the modules are periodically and
outside the scope of this paper, as we have focused on simultaneously recorded by the data-acquisition system
illuminated forward-bias I-V data. in approximately 300 ms, and one curve contains about
4000 data points. This high number of data points allows
for efficient filtering of noise and achieving sufficient
4 DETAILS OF THE EXPERIMENTAL SETUP resolution in the higher-current region. The sweep time
must be low since any variation in the irradiance during
4.1 Site and module details the sweep may lead to a variable photocurrent. A much
The 3 modules tested are polycrystalline-Si modules shorter sweep time than that used here, would have been
consisting of NS=60 screen-printed, series-connected desirable. Crystalline-Si modules allow for sweep times
cells of size 15.6x15.6 cm2. 5 bypass diodes are used, down to a few milliseconds [21]. However, the much
each connected across a substring of 12 cells. One of the higher sweep time of 300 ms was imposed by the
modules is based on SoG-Si produced from Elkem hardware and software used in our setup, as well as by
Solar’s metallurgical route (denoted Elkem Solar Silicon, the high number of data points we used for one I-V
ESS™ by the manufacturer), and the other two are made curve.
using classical polycrystalline-Si purified by the Siemens

4158
25th European Photovoltaic Solar Energy Conference and Exhibition /
5th World Conference on Photovoltaic Energy Conversion, 6-10 September 2010, Valencia, Spain

4.3 Equivalent Cell Temperature (ECT) In this Section we examine the behaviour at higher
Direct temperature measurements were performed on voltages of the logarithmic function described above. We
the module denoted A10156 via a Pt100 probe placed on create semi-logarithmic plots from empirical I-V data
the back surface metal contact of a central cell. In sets of the modules using different values for RS, and
addition, we calculated the Equivalent Cell Temperatures observe how the curves change near open circuit.
(ECTs) of all the modules during the I-V curve sweeps.
This was done after the procedure given in IEC 904-
5:1993 [22] using the in-plane irradiance, the open-
circuit voltage VOC and its temperature coefficient β
(available from an independent lab), and the value of
qe/(n1kBT) extracted from semi-logarithmic plots. In fact,
this IEC procedure assumes a one-diode model, which is
consistent with an approximation we make later in
Section 5.1, where we disregard the recombination-diode
term at higher voltages. The procedure is only applicable
to non-shaded I-V curves taken at irradiance above 200
W/m2. Our analysis was done on I-V data sets taken at
irradiance levels above 400 W/m2.
The readings from the Pt100 sensor were
systematically higher (by a few degrees) as compared to
ECT values obtained for the module denoted A10156.
Figure 3: Semi-logarithmic plots without compensation
Analysis of data taken at conditions very close to
for series-resistance effects (RS set to 0 Ω).
Nominal Operating Conditions (NOC) showed a typical
reading of the Pt100 sensor of about 46°C, whereas ECT
was about 42°C. Both values are in the range of typical
NOC temperatures found in the literature for glass-plastic
crystalline-Si PV modules [23,24]. However, we
considered ECT to be the more reliable estimate of p-n
junction temperature as its derivation is based on an
established international standard procedure and on
detailed module data from an independent, certified lab.

5 NEW APPROACH TO SEMI-LOG PLOTS AND RS

5.1 The logarithmic function


When studying the higher-voltage part of an I-V
curve, one can re-write Eq. (2) as shown below:
Figure 4: Semi-logarithmic plots with RS set to 0.80 Ω
V + IRS V + IRS (an exaggerated value) for all modules.
V + IRS N S n1VT N S n2VT
I PH − I = + I 01e + I 02 e (3)
RP

As n2 is close to double the value of n1 [2,15] or more


[4], the first exponent in Eq. (3) eventually dominates at
higher voltages over the other two terms on the right-
hand side. Dividing both sides by IPH makes the equation
dimensionless. This allows for taking natural logarithms
on both sides. The high-voltage approximation then gives

I −I   I 01  V + IRS
ln  PH  ≈ ln  + (4)
 I PH   I PH  N S n1VT
If the value of RS is known and n1 is constant with
Figure 5: Semi-logarithmic plots with RS set to 0.40 Ω
I −I  V + IRS for all modules. Nearly perfect linearity is achieved at
bias, the plot of ln  PH  versus will higher voltages.
 I PH  NS
be linear at higher voltages, with a slope equal to An erroneous value of RS results in a semi-
qe/(n1kBT) from which n1 can be deduced. The vertical- logarithmic plot deviating from linearity at higher
axis intercept of the linear fit will equal ln(I01/IPH) from voltages, as can be observed in Figs. 3 and 4. From a
which I01 is easily derived. given I-V curve measurement, a proper algorithm can
extract the value of the series resistance that results in the
5.2 Semi-log plots from experimental I-V data best linearity of the corresponding semi-logarithmic plot

4159
25th European Photovoltaic Solar Energy Conference and Exhibition /
5th World Conference on Photovoltaic Energy Conversion, 6-10 September 2010, Valencia, Spain

at higher voltages (see Fig. 5). As discussed in Sections


3.2 and 5.1, the slope of the linear region gives n1, 6 RESULTS AND DISCUSSION
whereas linear extrapolation to the vertical-axis intercept
gives I01. In this Section we present and analyze the values of
the parameters obtained from I-V curves recorded on a
5.3 The RS-estimation algorithm clear-sky summer day, 27 June 2010. We also introduce
For each I-V curve measured, the value extracted for a modified definition of the local ideality factor and
RS is the one that gives the best linearity in the demonstrate an m-V plot that is compensated for the
corresponding semi-logarithmic plot. Fig. 6 shows a plot series-resistance effect.
of the least-squares residual as a function of the value
chosen for RS for an individual I-V data set. We assume 6.1 Series resistance
that the minimum of this function represents the best
estimate of the module series resistance from the
particular I-V curve measurement.

Figure 7: Series resistance over cell temperature. Data is


from 27 June 2010, a clear-sky day.

Figure 6: Least-squares residual of a linear fit as a


function of the RS-compensation value used for an
individual I-V data set. The curve has a minimum which
represents the best estimate of RS from this particular I-V
curve measurement.

We developed an algorithm to find this value for


each measured I-V data set. A starting range of RS-values
between 0 and 1 Ω is used. (A higher upper limit may be
needed for modules of higher series resistances.) For 8
equidistant values in this range, RS-compensated semi-
logarithmic plots are created, and a linear fit is applied to
the high-voltage part of each plot. The resulting least-
squares residuals are then compared. Depending on
where the least-squares residual is minimal, a new,
Figure 8: Equivalent Cell Temperatures (ECTs)
narrower range of RS-values is selected for the next
evolution in the course of the sunny day.
iteration.
This procedure is repeated several times until an
accuracy of 0.1% or better is achieved. As will be seen in
Section 6.1, the actual variation in RS-values found for
consecutive I-V curves is much higher, due to e.g. noise
and irradiance instability during I-V sweeps.
When doing partial linear fitting of semi-logarithmic
plots, a proper choice of the low voltage limit is
important (see Fig. 5). Our study shows that the RS-
estimate for an I-V data set can vary by several per cent
depending on the limit chosen. This variation differs for
the different I-V data sets, even for consecutive ones. In
the following Section we present results obtained when
the limit is set to approximately 20% of ISC. This limit
has been found to be a sensible choice based on
numerical experiments on the data. However, we are
searching for a more robust criterion since the correct Figure 9: Evolution of the plane-of-module irradiance in
estimation of the remaining parameters depends strongly the course of the sunny day from which are the data
on the value found for the series resistance. analyzed. Values above 400 W/m2 are shown.

4160
25th European Photovoltaic Solar Energy Conference and Exhibition /
5th World Conference on Photovoltaic Energy Conversion, 6-10 September 2010, Valencia, Spain

Fig. 7 shows the dependence of the series resistance minority-carrier lifetime in the bulk of the Si wafer,
on cell temperature (ECT) for the modules denoted resulting in p-n-junction non-ideality and reduced fill
A10156 and A10160. Fig. 8 presents the evolution of factors [27].
ECTs in the course of the day, and Fig. 9 shows the
plane-of-module irradiance profile.
Minor variations in RS are seen during the course of
the day, and there is no practical difference between the
measured modules. (For clarity, data for the third module
is not shown on the plots, but the statement is valid for
that one as well.) A slightly upward trend with
temperature can be seen. The outlier and the spreads
visible in the results on Fig. 7 may be due to noise and/or
irradiance instability during the I-V sweeps, given the
relatively long sweep time of approximately 300 ms. In
order to minimize the effect on other parameters of
outliers and spread in the values obtained for RS, a
‘Smoothing Spline’ interpolation of RS versus time was
applied in Matlab. The smoothed values were then used
in the further estimation of n1 and I01. The accuracy in
Figure 11: Diffusion-diode ideality factor n1 over cell
the estimation of these parameters is very much
temperature. Data is from 27 June 2010.
influenced by RS.
We have verified the results obtained for the series
6.2.1 Local ideality factor analysis (m-V plots)
resistance by comparing them to alternative estimates
In this Section we propose what we believe to be an
derived by a conductance method [25] adapted for PV
improvement of the classical local ideality factor analysis
devices. For I-V curves of good quality, the two methods
(m-V plots). A modification of the classical definition is
give results for RS that differ by less than 0.5%.
introduced which compensates for the series-resistance
It should be noted that in a previous paper, where we
effect at higher voltages. As a consequence, the
focused on the development of the method and the
compensated m-V plots show asymptotic behaviour at
algorithm for RS-estimation, we already indicated the
higher voltages and provide a good approximation to n1.
ranges of n1 and I01 [1]. However, we later discovered an
Classical m-V plots may not be useful for obtaining
error in our computer code that lead to partially
the diffusion-diode ideality factor, as they are not
erroneous estimations of the parameters. The computer
compensated for the series-resistance effect at higher
code has now been corrected, but since the method is still
voltages, whereas various detrimental effects take place
under development, the values obtained for the
at lower to moderate voltages [15]. Thus, m-V curves
parameters must be considered preliminary.
from empirical data often lack regions of constant m
corresponding to a single exponential conduction
6.2 Ideality factors
mechanism (like the diffusion current across the p-n
junction). Below, we propose a modification to the
classical m-V plot which eliminates the effect of RS at
higher voltages. Without this modification, the (classical)
local ideality factor increases at higher voltages. In
contrast, a RS-compensated m-V plot can have a flat
region with a constant local ideality factor, as is shown
below.
The local ideality factor of a PV cell in the dark is
given by Eq. (5) [15].
I  dV 
m=   (5)
VT  dI 
For its calculation, the inverse slope at each point of the
I-V curve must be calculated. In the case of illuminated
characteristics, the current I must be substituted by ISC–I
Figure 10: Evolution of the diffusion-diode ideality (cf. Eq. (3)):
factor n1 in the course of a sunny day.
m=
( I SC − I )  dV 
Fig. 10 shows the evolution of the diffusion-diode   (6)
ideality factor, n1, for two of the modules. Plots of n1 VT  d ( I SC − I ) 
over ECT are shown in Fig. 11. (The module denoted We modify this definition by adding a series-
A10156 is the one based on SoG-Si produced from a resistance compensation term (IRS) to the voltage V. By
metallurgical route.) A simple correlation with cell taking the number of cells in series (NS) into account, the
temperature or irradiance cannot be inferred at this stage. expression becomes valid also for PV modules:
Ideality factors higher than 1 are typical for
polycrystalline-Si cells and have been traced to the
mSR =
( I SC − I )  d (V + IRS ) 
  (7)
presence of metallic impurities [26]. The latter have been N SVT  d ( I SC − I ) 
found to cause injection-level dependence of the

4161
25th European Photovoltaic Solar Energy Conference and Exhibition /
5th World Conference on Photovoltaic Energy Conversion, 6-10 September 2010, Valencia, Spain

Fig. 12 shows a plot of the RS-compensated local The preliminary results are in good but not perfect
ideality factor mSR created from an I-V data set of the correlation with the theoretical model; there is a certain
module denoted A10160. As can be seen, a practically spread in the estimated data and there is a clear deviation
constant level of the local ideality factor is reached at from the theoretical model for the module denoted
higher voltages. It can be shown that this value is nearly A10160. Nevertheless, it is clear that the module denoted
equal to the diffusion-diode ideality factor n1 as it is A10156 (the one based on SoG-Si produced from a
found by the method elaborated in Section 5.1, if both metallurgical route) has lower values of I01 (B≈1.7x103
approaches are applied to the same I-V data set. AK-3 versus B≈3.5x103 AK-3 for the module denoted
The characteristic ‘humps’ at moderate voltages are A10160). This may indicate a better quality of the p-n
usually attributed to the combined effects of the shunt junctions implemented in its cells. This also explains the
current through RP and one or more recombination higher STC-value of the open-circuit voltage VOC
mechanisms [15], and are not analyzed in this paper. compared to the other two modules based on classical
polycrystalline-Si.

7 CONCLUSION

We have presented a new method for extracting the


series resistance from individual I-V curves of
crystalline-Si PV modules. It facilitates the further
extraction of parameters of the two-diode cell model that
we have adapted for modules. Our approach is based on
linear fitting of data and data array manipulations, and
avoids the ambiguity and complexity of the non-linear
curve fitting approaches. It can achieve high accuracy,
but is limited by the quality of the measured I-V data.
The method was applied to outdoor I-V data from
Figure 12: A RS-compensated m-V plot of the module modules based on two types of polycrystalline SoG-Si.
denoted A10160. The results indicate that each module has cells that are
nearly identical so that the 7-parameter, double-
6.3 Reverse saturation current I01 exponential model is indeed applicable for the modules.
Fig. 13 shows plots of the reverse saturation current Activation of bypass diodes at lower voltages can prevent
over ECT for two of the modules. The discrete points the estimation of the parallel resistance and influences
represent estimates from experimental data, whereas the the estimate of the photocurrent from illuminated I-V
continuous lines are plotted on the basis of the theoretical curves. Nevertheless, the method of analysis presented in
temperature dependence given by Eq. (8) [7]. this paper gives valuable information about the
performance of PV modules, and enables a comparison
 E  between modules.
I 01 (T ) = BT 3 exp  − G  (8) We have also introduced a new definition of the local
 k BT  ideality factor featuring compensation for the series-
resistance effect. It has been shown that for
polycrystalline-Si modules such an m-V plot reaches a
nearly constant value of the local ideality factor at higher
voltages. This value is practically equal to the ideality
factor found from the corresponding semi-logarithmic
plot.

8 ACKNOWLEDGEMENT

The authors acknowledge financial support by Elkem


Solar, the Research Council of Norway and the
Municipality of Kristiansand. We further thank Torfinn
Buseth and Jan Ove Odden at Elkem Solar for technical
expertise, valuable discussions, and sharing of PV
module reference data as measured by the independent
Figure 13: Diffusion-diode reverse saturation current I01 laboratory in Germany. Special thanks are due to Erling
over cell temperature for two of the polycrystalline-Si Andresen as well as the technical staff and the students
modules tested (discrete points). Data is from the sunny from Kvadraturen Skolesenter in Kristiansand. They
day of 27 June 2010, taken at irradiances above 400 have contributed in an important way to the project by
W/m2. The module denoted A10156 is the one based on building the mechanical frames and support structures for
SoG-Si produced from a metallurgical route. The the PV modules, and by doing the electrical cabling, all
continuous lines are plotted on the basis of the theoretical done very professionally. We also thank Prof. Hans
temperature dependence given by Eq. (8). Georg Beyer for reviewing the paper and for sharing his
experience and knowledge with us.

4162
25th European Photovoltaic Solar Energy Conference and Exhibition /
5th World Conference on Photovoltaic Energy Conversion, 6-10 September 2010, Valencia, Spain

illuminated current-voltage characteristics of solar


cells”, IEEE Transactions on Electron Devices 33
9 REFERENCES (1986) 391.
[17] J. Carstensen et al., “Characterization of the grid
[1] G.H. Yordanov, O.M. Midtgård, and T.O. Saetre, design by fitting of the distributed serial grid
“Extracting parameters from semi-log plots of resistance to CELLO resistance maps and global I-V
polycrystalline Silicon PV modules I-V data: curves”, Proceedings 24th European Photovoltaic
Double-exponential model revisited”, presented at Solar Energy Conference (2009) 516.
the 35th IEEE Photovoltaic Specialists Conference [18] B.G. Streetman and S. Banerjee, “Solid State
(2010). Electronic Devices”, 5th Edition, Prentice Hall
[2] J. Appelbaum, A. Chait, and D. Thompson, (2000) Chapter 5.6.3 “Ohmic Losses”, 217.
“Parameter estimation and screening of solar cells”, [19] S.M. Sze and K.K. Ng, “Physics of Semiconductor
Progress in Photovoltaics: Research and Applications Devices”, 3rd Edition, John Wiley & Sons (2007)
1 (1993) 93. Chapter 2 “p-n Junctions”, 79.
[3] R. Gottschalg et al., “Comparison of different [20] D. Pysch, A. Mette, and S.W. Glunz, “A review and
methods for the parameter determination of the solar comparison of different methods to determine the
cell double exponential equation”, Proceedings 14th series resistance of solar cells”, Solar Energy
European Photovoltaic Solar Energy Conference, Materials & Solar Cells 91 (2007) 1698.
Vol. I (1997), 321. [21] A. Virtuani et al., “Comparison of indoor and
[4] M. Wolf and H. Rauschenbach, “Series resistance outdoor performance measurements of recent
effects on solar cell measurements”, Advanced commercially available solar modules”, Proceedings
Energy Conversion 3 (1963), 455. 23rd European Photovoltaic Solar Energy Conference
[5] H. Müllejans et al., “Reliability of the routine 2-diode (2008) 2713.
model fitting of PV modules”, Proceedings 19th [22] IEC, International Standard 904-5, 1st Edition
European Photovoltaic Solar Energy Conference (1993).
(2004), 2459. [23] A. Luque and S. Hegedus, editors. “Handbook of
[6] J. Hyvärinen and J. Karila, “New analysis method for photovoltaic science and engineering”, Wiley, ISBN
crystalline silicon cells”, Proceedings 3rd World 0-471-49196-9 (2003), Chapter 20.10 “PV generator
Conference on Photovoltaic Energy Conversion behaviour under real operation conditions”, 947.
(2003) 1521. [24] O.M. Midtgård et al., “A qualitative examination of
[7] S. Coors and M. Böhm, “Validation and comparison performance and energy yield of photovoltaic
of curve correction procedures for silicon solar modules in southern Norway”, Renewable Energy 35
cells”, Proceedings 14th European Photovoltaic Solar (2010) 1266.
Energy Conference, Vol. I (1997), 220. [25] J.H. Werner, “Schottky barrier and pn-junction I/V
[8] J.-P. Charles et al., “Consistency of the double plots – small signal evaluation”, Applied Physics A
exponential model with the physical mechanisms of 47 (1988) 291.
conduction for a solar cell under illumination”, [26] A. Cuevas, “Lifetime studies of multicrystalline
Journal of Physics D: Applied Physics 18 (1985), silicon”, Presented at the 8th Workshop on Crystalline
2261. Silicon Solar Cells Materials and Processes (1998).
[9] J.J. Loferski, “The first forty years: A brief history of [27] D. Macdonald and A. Cuevas, “Reduced fill factors
the modern photovoltaic age”, Progress in in multicrystalline silicon solar cells due to injection-
Photovoltaics: Research and Applications 1 (1993), level dependent bulk recombination lifetimes”,
67. Progress in Photovoltaics: Research and Applications
[10] A. Goetzberger, C. Hebling, and H.-W. Schock, 8 (2000) 363.
“Photovoltaic materials, history, status and outlook”,
Materials Science and Engineering 40 (2003) 1.
[11] W. Shockley, “The theory of p-n junctions in
semiconductors and p-n junction transistors”, Bell
Systems Technical Journal 28 (1949), 435.
[12] B.G. Streetman and S. Banerjee, “Solid State
Electronic Devices”, 5th Edition, Prentice Hall
(2000) Chapter 5.6.2 “Recombination and generation
in the transition region”, 214.
[13] Z. Ouennoughi and M. Chegaar, “A simpler method
for extracting solar cell parameters using the
conductance method”, Solid State Electronics 43
(1999) 1985.
[14] V. Lo Brano et al., “An improved five-parameter
model for photovoltaic modules”, Solar Energy
Materials & Solar Cells 94 (2010) 1358.
[15] K.R. McIntosh, “Lumps, humps and bumps: Three
detrimental effects in the current-voltage curve of
silicon solar cells”, PhD Thesis, University of New
South Wales (2001).
[16] G.L. Araújo, A. Cuevas, and J.M. Ruiz, “The effect
of distributed series resistance on the dark and

4163

Anda mungkin juga menyukai