Presented at the PVSC34 Philadelphia 11 Jun 2009

UNDERSTANDING kWh/kWp BY COMPARING MEASURED DATA WITH

MODELLING PREDICTIONS AND PERFORMANCE CLAIMS

Steve Ransome (SRCL) UK

mailto:steve@steveransome.com Tel +44 (0)7515 565 010 http://www.steveransome.com

ABSTRACT width of module bins (e.g. a 200W nominal module bin

may contain modules from 200 to 210Wp).
The performances of PV arrays are usually
compared by the kWh/kWp produced. Several recent If specific module technologies or manufacturers
independent studies in European climates [1][2][3][4] really did have large differences in energy yields then
have found similar kWh/kWp (within experimental these results should be repeatable and measurable by
error ±5%) for different module technologies without a all studies.
systematic bias in favour of any technology.
Sizing programs that estimate kWh/kWp values UNCERTAINTIES OF kWh/kWp STUDIES
usually predict the IV curve of a PV module with a 1-
diode model [5] to match the STC curve then correct Outdoor yield results are often given without
for irradiance and temperature effects. They often quoting any inaccuracies or how any corrections are
predict higher kWh/kWp for thin film devices as their made for downtime, faulty or out of spec data or
models give them better low light level performance shading, Vmax mistracking etc.
than is measured. If a module X is found to have a higher energy
Measured weather parameters such as tilted plane yield than a module Y then will these differences be
irradiance and detailed dc performance of several applicable to just those two modules, all modules of
technologies at ISET has been compared and those types, all modules made by the manufacturers
contrasted with predictions from three different sizing or all modules of the technologies involved ?
programs. Recent surveys have shown differences of
Many differences have been found between kWh/kWp are less than the earliest measurements - it
measured and modelled weather data, thermal is probably due to better Rshunt performance which
coefficients and low light level dependency meaning has raised low light level on c-Si and thin films, more
kWh/kWp should not vary much between correctly accurate Pmax definitions and lower allowance for TF
declared PV technologies. degradation.
If kWh/kWp variations are small (i.e. <±5%) shown Attempts to measure the real power in the field
in Europe then other parameters have been using translations for temperature and irradiance have
suggested that can be used to differentiate PV limited accuracy (as proven by some round robin
technologies. tests)- particularly for multi junction thin film devices
with their thermal annealing, also for recent high
kWh/kWp PERFORMANCE CLAIMS power c-Si with their high capacitances.
Indoor measurements to show dependence of
Some manufacturers claim "values of up to 30% efficiency vs. light level etc are often incomplete as
higher kWh/kWp" for their products due to better outdoors effects are correlated (for example as the
performance under some or all of the following light level rises the temperature will necessarily
weather conditions :- increase).
low light levels Tracking modules away from the sun to measure
high diffuse light fraction low light levels causes inaccuracies as these show
non optimal tilt (e.g. horizontal mounting) higher angle of incidence reflectance effects at blue
high temperatures rich spectra that will never be achieved under these
bluer light (summer clear sky) conditions.
thermal annealing (e.g. autumn will be better than
spring) kWh/kWp PREDICTIONS FROM MODELLING
“foggy weather” PROGRAMS

Often their published measurements appear to Sizing programs will usually perform the following
show better yields from their products than other steps to estimate kWh/kWp from a PV array
competitors’ technologies that they have measured. Calculate stochastic tilted plane hourly irradiance
However due to instabilities of Pmax (initial series from monthly horizontal plane insolation.
stabilisation, steady decline and changes due to Estimate module temperature from NOCT value.
thermal annealing and spectral mismatch) it is not Generate hourly dc Pmax as a function of
known how their nominal STC values have been Irradiance and temperature usually with a 1-diode
declared. model [5].
Previous publications [1][6][7] have shown that the Estimate other losses such as soiling, shading,
yields of systems may vary by ±4-5% just due to and mismatch.
uncertainties in reference module calibrations and the Determine inverter efficiency, Vmax tracking and
wiring losses.
 Presented at the PVSC34 Philadelphia 11 Jun 2009

Results: sum over a year to get kWh/kWp. Figure 3 gives the total insolation measured vs.
irradiance (y-axis) and Tmodule bins (x-axis) for a c-Si
The programs usually treat weather data as with the weather data from figure 2.
independent variables but all weather parameters are The peak insolations occur around 900W/m² and
correlated [6]. Figure 1 illustrates module temperature Tmodule=45C.
vs. air mass for clear skies in Kassel. A correlation is Figure 4 shows the percentage change in
seen from the slope of the trend line such that any insolation at each irradiance and module temperature
attempt to extract a coefficient (for example of module bin when the 15 second data is averaged to hourly.
temperature) will have a spectral dependence to be The bins around 800W/m² irradiance and 40C
considered. Tmodule fall dramatically (by 10 to 50%) and the
apparent irradiance bins on a line from 200W/m² @
20C to 800W/m² @ 50C are increased.
This will affect energy yield predictions somewhat
for modules that are modelled not to have a constant
efficiency with irradiance and or temperature (whether
or not this is true in reality).

Figure 1: Plot of module temperature vs. air mass for

clear sky conditions for a c-Si module in Kassel.

PLANE OF ARRAY INSOLATION vs. IRRADIANCE

It has been found previously [8] that averaging

high frequency meteorological data to hourly will
change the distribution of the irradiance vs. insolation
curve as there will be variable weather hours (some
partly sunny, other partly cloudy conditions) that will
be averaged to be a “dull hour”.
Figure 2 gives the percentage of insolation Figure 3. Distribution of Figure 4. Change in
measured at each irradiance bin when measured at plane of array insolation insolation (% of
15 second intervals in Kassel for 2003. When vs. module temperature kWh/m²/y) per Tmodule
averaged to hourly values there is a steadily rising and Irradiance measured and irradiance bin from
amount of irradiance to 900W/m² then a fast drop to every 15 seconds at 15 sec to averaged 1
zero at 1100W/m². Increasing the sampling frequency ISET hour values.
to every 10minutes, every 1 minute and finally every
15 seconds shows an increasing amount or insolation
occurring at higher irradiances – up to 1350W/m² for MEASURED vs. MODELLED PV EFFICIENCY vs.
brief periods for example when there might be direct LIGHT LEVEL
sun in a white sky where the diffuse reflection was
higher than expected. The measured performance of 7 PV modules was
compared with the model in Sizing programs’
databases as listed in Tables 1 and 2.

Table 1: PV module technologies tested

PV Modules Technology
(not in same order as module #)
c-Si mc-Si, high performance and
#1, #2, #3 standard sc-Si
TF Thin Film
#4, #5, #6, #7 (1J and 3J a-Si, CIS and CdTe)

Table 2: Sizing programs used

Sizing program Comment
X Commercial or freely
Y available Sizing programs
Figure 2. In plane insolation measured at 15second
Z
intervals at Kassel, Germany and averaged at
different frequencies down to hourly.
Figure 5 illustrates the measured relative efficiency
vs. irradiance and module temperature for a thin film
module measured in Germany (top left) and that
 Presented at the PVSC34 Philadelphia 11 Jun 2009

predicted for three different sizing programs X, Y and

Z. (1) Pmax @ STC (1000 W/m², Tmodule=25C)
The measured module (top left) had degraded (2) Gamma (1/Pmax * dPmax/dT) @ 1000W/m²
below its nameplate value. It would have experienced (3) Eff @ NOCT (800W/m² and Tmodule ~47C)
all of the weather parameters in Figure 5 correlated (4) Efficiency reduction at low light levels: 200W/m² /
with the irradiance, for example the module 1000W/m² (EN50380)
temperature of 55C has a higher relative performance
than other temperatures presumably because at high Not all manufacturers are yet quoting values (3)
temperatures the solar height will be tend to be and (4) (according to EN50380) but (1) and (2) are
higher, the air mass lower and hence the spectrum available.
bluer plus any thermal annealing will have come into
effect.
The models used by the three sizing programs X,
Y and Z are not only different from the measured data
but different from each other. X has a falling efficiency
with light level then a small peak, Y is flat and Z is
rising all the way as the irradiance falls.

Figure 6 shows similar graphs for a mc-Si module

and its nearest equivalent in the three Sizing
programs’ databases.
There is a little more similarity between the module
and the models but programs X and Y have a worse
low light response than expected.
Figure 7: Checking a Sizing program’s derived
efficiency parameters with those on a manufacturer’s
spec sheets.

A random study of several different modules from

their latest spec sheets and web data with the
different programs gave varying values of the 3 values
(2) to (4), almost all differing from the manufacturers’
published data which will affect the predicted energy
yield.

Table 3: Manufacturers data vs. modelled

assumptions for some present PV modules
TF TF c-Si c-Si
Figure 5. Thin Film module #7: Normalised efficiency
(i) (ii) (iii) (iv)
vs. irradiance and average module temperature (10,
25, 40, 55 and 70C) Clockwise from top left: (2) Gamma ( = 1/Pmax * dPmax/dT)
Measured, X, Z, Y. (manuf) -.33% -.21% -.45% -.34%
(models) -.25% -.24% -.43% -.38%
-.27% -.25% -.58% -.46%
(3) Eff@NOCT/ EffSTC
(manuf) 93.8% 96.3% 90.9% 91.7%
(models) 91.1 92.2% 86.5% 89.1%
95.4% 95.8% 90% 90.8%
(4) Eff200/ EffSTC
(manuf) 102% n/k 98.6% n/k
(models) 91.6% 94.2% 92.9%
102% 100.8% 82.6%

ESTIMATING THE EFFECT OF EFFICIENCY

MODELLING ON kWh/kWp DATA
Figure 6. c-Si module #1: Normalised efficiency vs.
irradiance and average module temperature (10, 25, Sizing programs essentially calculate kWh/kWp by
40, 55 and 70C) Clockwise from top l: Measured, X, Z, multiplying the hourly plane of array irradiance by the
Y. efficiency at the given weather conditions (irradiance
and module temperature. This can be summed by
INCONSISTENCIES FOUND IN SOME OF THE multiplying the insolation * efficiency by module
MODELS temperature and irradiance bin. Figure 8 plots
surfaces of insolation and efficiency vs. Tmodule and
Modelling of the efficiency vs. irradiance and irradiance, two values are marked for a cool low light
module temperature should agree with the and a warm high light condition to illustrate how these
measurement data on the manufacturers’ spec should be multiplied and summed together.
sheets. The following four values are illustrated in
figure 7.
 Presented at the PVSC34 Philadelphia 11 Jun 2009

by modelling data from Sizing programs X, Y and Z for

Kassel with measured and averaged to hourly (3600),
Insolation vs. Tmodule Modelled Efficiency vs. 5 minute (300) and 15 second data vs. hourly
and Irradiance Tmodule and Irradiance modelling data

MEASURED EFFICIENCY vs. WEATHER

PARAMETERS

The relative efficiency vs. light level for various

meteorological parameters was measured and plotted
Figure 8: Illustration of kWh/kWp summing by for four different module technologies – two c-Si and
multiplying insolation * predicted efficiency at each two thin films and is shown in figure 11 - each quarter
Tmodule and irradiance bin section shows the relative performance from (60 to
100% of nominal) vs. Irradiance (top), Beam Fraction
kWh/kWp results are given in Figure 9 for sizing (centre) and Angle of incidence (bottom). In each case
program X. Differences of up to 9% favouring the thin the lines are remarkably similar to one another, there
film module #7 over the c-Si module #1 for all three are no steep rises nor falls under extreme weather
sites from the modelled data. conditions that differentiate any of the technologies
from each other that could give rise to large
differences in kWh/kWp (assuming the
Pmax.actual/Pmax.nominal are similar).

Figure 9: Predicted performance ratio check from

sizing program X.

Predictions for kWh/kWp have been made with (i) IRRADIANCE

Met data from Kassel – both from an hourly generator
and also from 15 second averaged to 5 minutes and 1
hour. The overall plane of array insolation for the
hourly model was scaled to the same value as that
from the measured data for a fair comparison. Figure
10 illustrates the data looking for changes for each
model. All of them predict a slightly higher relative
performance ratio (less than 1%) from the 15 second
data than the hourly for both c-Si 1 and TF 6 but the
hourly model data is predicted to give a 1% higher
energy yield for all models + programs except for 6Z
which is around 3% as its model gives a much higher
efficiency at low light level than occurs.
(ii) BEAM FRACTION – Diffuse (left) to Direct (right)

(iii) ANGLE OF INCIDENCE – Normal incidence right

Figure 11 : Performance factor of four modules vs.

Figure 10 : Estimated relative performance ratio for a Irradiance, Beam fraction and angle of incidence
c-Si module #1 and a Thin Film module #6 calculated
 Presented at the PVSC34 Philadelphia 11 Jun 2009

WHICH OTHER PARAMETERS CAN Difficulties in the definitions of kWp will be

DIFFERENTIATE TECHNOLOGIES ? experienced when comparing technologies that are
not monofacial and planar – for example curved or
If kWh/kWp expected from correctly declared and bifacial devices or passive or active trackers will have
stable (or nearly) technologies really are within a different apparent Wp from what a standard flash
measurement accuracies (±5%) then other tester will measure.
parameters will need to be used to distinguish Some other options to differentiate technologies
between different technologies. are are listed below in table 4.

Table 3 : Other parameters that can differentiate pv technologies better than kWh/kWp

Technology High Efficiency Standard Thin Film Organics/Plastic

c-Si mc-Si a-Si, CIGS, CdTe
kWh/kWp Tests in Europe show often within ± 4-5% for correctly labelled, non n/k
degrading modules
W/m² @ STC 170-200 140-170 50-110 n/k
$/Wp (will vary) Should be very low
Solar Buzz - Jan 2009 4.00USD 3.05EUR 3.99USD 3.05EUR 3.27USD 2.50EUR
Photon - Apr 2009 [9] 3.17USD 2.40EUR 2.64USD 2.00EUR
Lifetime to >80% 20-25y guarantee 20-25y guarantee ~20-25y guarantee n/k
Initial Power (see
guarantee)
Flexible substrates – No No Some Yes
are they necessary ?
Visual Appearance ~Squares (with bus bars except back contact) Monolithic with narrow parallel cuts
Aesthetics Multi crystalline may be more reflective
Coloured back sheet or transparent glass
Shade tolerance May be worse as squarer cells May be better with high aspect ratio cells
if oriented with cells perpendicular to
shading
kg/Wp (no structure) (1 glass) (1 glass) (2 glass)
(Framed) 0.06-0.08 0.07-0.09 0.16-0.22 n/k
(Frameless) ~0.16
(Flexible) 0.06
kg/m² (no structure) (1 glass) (1 glass) (2 glass) n/k
(Framed) ~12 ~12 ~17
(Frameless) ~17
(Flexible) ~4
Temperature coefficient ~-0.35%/K ~-0.45%/K ~-0.25%/K n/k
Large spectral No No Yes, particularly Yes
difference with. multijunction
Pyranometer ?
Seasonal/Thermal No No Yes n/k
annealing ?
Restrictions in use ? Some technologies
may be banned in
certain countries
Ruggedness (ability to c-Si mostly laminated with glass and usually TF on plastic or metal foils may function
function after damage) unable to function after breakage even after being damaged
Energy Cost c/kWh To be confirmed – depends on $/Wp, Efficiency and longevity and Insolation

Sizing program models for efficiency vs. irradiance

CONCLUSIONS and module temperature can be very different from
measured data for c-Si and thin films, their
Weather data generators in sizing programs often kWh/kWp predictions depend critically on the
predicts more insolation at low light levels than curve shape
that which occurs Errors in the values of gamma
Real kWh/kWp measurements are often similar at (1/Pmax*dPmax/dT), Eff@NOCT and
least in Europe (within experimental error ±5%) for Eff@200/Eff@1000W/m² have been seen vs.
different module technologies when correctly manufacturers claims.
declared stable Pmax values are used. No effects have been found which could give
Studying the efficiency vs. irradiance, beam widely varying kWh/kWp values (i.e. ±30%
fraction, angle of incidence for thin films and claimed) between technologies as long as the
crystalline indicate less variation between the correct Wp at STC is achieved.
technologies than is claimed As kWh/kWp values do not differentiate various
PV technologies at least in Europe then other
 Presented at the PVSC34 Philadelphia 11 Jun 2009

parameters have been suggested that can be [3] “Temperature and intensity dependence of twelve
used instead. photovoltaic technologies” Zinßer et al Institut für
Further checks should be taken in more extreme Physikalische Elektronik, Universität Stuttgart,
climates [4] Photon Magazine “A new best module” Photon
International Feb 2009
ACKNOWLEDGEMENTS [5] “Improvement and validation of a model for
Peter Funtan and ISET, Kassel for the dc photovoltaic array performance “ De Soto, Klein and
measurements Beckman model Solar Energy, 80, 78-88, (Jan 2006).
[6] “How well do pv modelling algorithms really predict
REFERENCES performance?” S Ransome 22nd PVSEC Milan 2007
[7] "Modelling inaccuracies of PV energy yield
[1] "A Review of kWh/kWp Measurements, Analysis simulations" S Ransome 33rd PVSC 2008 San Diego
and Modelling" S Ransome Valencia 2008 23rd [8] “Why hourly averaged measurement data is
PVSEC insufficient to model PV system performance
[2] “Direct performance comparison of pv module” D. accurately” S Ransome and P Funtan 2005 Barcelona
Chianese et al ISAAC-TISO 20th PVSEC
[9] Photon International May 2009 p76