Conference PaperPDF Available

On-site performance verification to reduce yield prediction uncertainties

Authors:
  • Enmova GmbH

Abstract and Figures

In this paper, we describe a number of quality assurance procedures for PV performance evaluations using data that has been acquired with commercially operating PV power plants. Summarized under the term " performance verification " , these procedures aim to make reliable use of data gathered by PV operators, despite inaccuracies in such data. To this end, short-term measurements with independent equipment and/or data sources are factored in. This ensures both meteorological and PV performance data gathered by operators have sufficient quality. Given a sufficient quality, site-adaptation of satellite based solar irradiation time series and optimizations of PV models may allow for significantly reduced uncertainties of yield predictions. This we anticipate to be of great interest to a broad range of PV stakeholders. The prerequisite for this, however, are appropriate quality of on-site sensors and strict maintenance of the PV performance monitoring systems in place.
Content may be subject to copyright.
On-site performance verification to reduce yield prediction uncertainties
Nils Reich, Julian Zenke, Björn Müller, Klaus Kiefer, Boris Farnung
Fraunhofer ISE, Freiburg, D-79110 Germany
AbstractIn this paper, we describe a number of quality
assurance procedures for PV performance evaluations using data
that has been acquired with commercially operating PV power
plants. Summarized under the term “performance verification”,
these procedures aim to make reliable use of data gathered by PV
operators, despite inaccuracies in such data. To this end, short-
term measurements with independent equipment and/or data
sources are factored in. This ensures both meteorological and PV
performance data gathered by operators have sufficient quality.
Given a sufficient quality, site-adaptation of satellite based solar
irradiation time series and optimizations of PV models may allow
for significantly reduced uncertainties of yield predictions. This
we anticipate to be of great interest to a broad range of PV
stakeholders. The prerequisite for this, however, are appropriate
quality of on-site sensors and strict maintenance of the PV
performance monitoring systems in place.
Index Terms Performance, performance ratio, PR testing,
PR verification, bankability, financing, utility-scale PV.
I. INTRODUCTION
The performance ratio (PR, defined in IEC 61724) is a
quantity to indicate how well a PV system performs with
respect to actual irradiation and nominal module efficiency.
One may further distinguish a number of loss mechanisms
from both theoretical considerations (i.e., yield simulations)
and practical facts (i.e., field measurements). Comparing
expected and actual performances, valuable information on
whether the system performs as expected can be obtained [1].
Here, it is already common practice to compare PR related to
losses on direct-current (DC) and alternating-current (AC)
sides of a system, both of which can be measured relatively
easily and is simulated in any case (denoted PRDC and
PRAC). In addition, seasonal and inter-annual variations of
PR are analyzed in daily work routines of PV performance
analysts. Notwithstanding, no update of above noted IEC
seems expectable. To this end, it seems hard for the PV
modelling community at this moment to agree on common
notations for the various losses occurring in the energy loss
chain, and information on how these losses were calculated
are often omitted. Moreover, related uncertainties are often not
addressed either.
In this paper, we aim to summarize our modelling activities
directed at both yield simulations and PV monitoring data
validation. This may contribute in complementing existing
standards and may help to get more confidence of the fidelity
of the data recorded in operating and maintenance (O&M) of
commercial PV power plants.
For energy yield prediction studies, the particular system
case needs to be carefully reviewed and the given plant
documentation needs to me meticulously implemented into the
yield simulation model. Only then it is possible to state
realistic uncertainties of individual loss mechanisms involved.
Otherwise, one is hardly able to perform sensitivity analyses
on various loss factors involved, with the goal to determine
uncertainties of each loss mechanism to be at the lower,
middle or higher end for the particular system case. In order to
get an impression of individual uncertainties involved and
their theoretically possible reductions, Table I gives an
overview on the fairly broad ranges we currently perceive
reasonable for c-Si PV systems. As for the scope of these
figures, do note stated ranges relate to one standard deviation
(sigma=1) and are only indicative to standard c-Si technology,
aggregated annual yield of the first operational year of one
individual system, and may be somewhere between the noted
lower and upper end for a PV plant under investigation.
TABLE I
ORDERS OF MAGNITUDE OF UNCERTAINTY, SEE TEXT.
Source of uncertainty
Uncertainty range [%]
(one standard deviation)
individual combined
A: Irradiation
Satellite GHor 3…5
4…6
Direct/diffuse-
ratio and GPOA
transposition
2…4
B: Shading
Horizon 0…0.5 1…4
Inter-row 1…4
External 0…3 n/a
C: Soiling 1…4 n/a
D: Deviation
from STC
Reflection < 1 < 2
Spectral losses ~1
Irradiance
intensity
1…2
Temperature 1…2
E: Actual PV
capacity and
DC-losses
Module power
and mismatch
2…5 < 5
Cabling 0.5
F: Inverter
losses
Model related < 1.5 < 2
Power limitation < 0.5
G: AC-losses Cabling 0.5 < 1
Transformer 0.5
Combined total initial uncertainty
Typically 5 … 11%
978-1-4799-7944-8/15/$31.00 ©2015 IEEE
The overview of uncertainty ranges indicated in Table I
only provides the order of magnitude we suggest reasonable
for very different PV power plant cases, i.e., using a variety of
components each having different performance- and system-
related uncertainties, installed somewhere on earth, which also
involving uncertainty, predominantly in meteorological data
available for simulations.
Unless applied to a specific PV power plant case, listed
ranges of Table I have limited practical use. To this end, we
summarize in the following section a number of studies, ideas
and also practical modelling steps we currently use when
evaluating uncertainty sources and their potential reduction.
The letters A through G listed in Table I for each category
correspond to the following sub-chapters.
II. OPPORTUNITIES AND LIMITATIONS TO REDUCE
UNCERTAINTY
A. Actual on-site irradiation resource
It is clear from the outset of Table I that uncertainties of
category „A: Irradiation“ dominate combined uncertainty in
the initial phase of PV power plant operation. Both irradiation
on the horizontal plane and the inevitable first transposition
modelling step (calculating irradiation in plane-of-array
(GPOA)) have relatively high associated uncertainties.
Fortunately, a substantial reduction of these uncertainties may
get realized using on-site data of the first year of PV power
plant operation. Here, the big advantage of satellite derived
data is (once again) its general availability. In fact, one may
from the first day on compare on-site irradiation with satellite-
derived irradiation data and, using so-called site-adaptation
processes, significantly reduce this uncertainty source down to
+/- 2.5% for GHI and +/- 4% for DNI [2] after 12 month (one
year, thus covering seasonal effects). The prerequisite for this
reduction, however, is accurate on-site data. Unless irradiation
measured on-site is accurate and trustworthy, this approach is
pointless. Fraunhofer ISE therefore regularly performs
verification measurements of on-site equipment, as to grant
sensors and logger systems at the plant provide data well
within confidence levels needed. Figure 1 pictures an example
of such a temporary irradiation measurement.
Fig. 1: Example photo of short-term on-site measurements.
B. Shading
More accurate irradiation data implies less uncertainty for
direct/diffuse ratio. This already helps reducing uncertainty of
shading losses as calculated for yield prediction reports: Using
performance data acquired on-site, one can compare the more
accurate direct/diffuse ratios to shading effects as they appear
in the performance data. To analyze, if shading loss models of
yield simulations accurately reflect particular inter-row
shading conditions, one may investigate PR as a function of
sun position. This way, the characteristic profile angle of a site
can be determined. The profile angle is defined as the
projection of the solar elevation angle onto a vertical plane
orthogonal to the azimuth of the module rows. An example is
shown in Fig. 3. Here, a profile angle of 15° can be inferred.
More details are given elsewhere [3].
Fig. 2: PR as a function of profile angle [3].
C. Soiling
Despite being easy to understand in practical terms, soiling
aspects can impose great difficulties when detailed predictions
are needed. A tendency for low soiling in geographical
locations with frequent rainfall on the one hand is in stark
contrast with locations having high dust accumulation
potential and infrequent rain events. For the latter, losses often
become so high that soiling losses (and related uncertainties)
in practical terms will be limited only by cleaning schedules
and appropriate O&M implementation: In many arid regions,
yield and related economic loss would become unacceptably
high without appropriate cleaning procedures.
At best, site-specific soiling rates can be inferred from on-
site monitoring data, if dedicated measurement installations
are deployed to quantify these losses. However, as of yet such
soiling measurement systems are not widely used, and
although their benefit is likely to be high, there has so far been
little data on how representative one soiling measurement
station is for a multi MW plant. In practice, one has often to
rely on noted dates that have been logged for cleaning events.
978-1-4799-7944-8/15/$31.00 ©2015 IEEE
In this case, one may obtain the increase of performance, due
to the cleaning event, from good-quality data of the on-site
monitoring.
There are a number of different types of soiling. Here, we
give an example of non-uniform shading induced by algae
soiling at the module edge (see upper pane in Fig. 3).
Fig. 3: Soiling by algae (upper pane) and dust (lower pane).
Fig. 4: Removing algae, example of DC-voltage decrease.
Fig. 5: DC-voltage as a function of module temperature.
In the soiling example illustrated in Figs. 3 through 5, lichen
and algae imposing inhomogeneous shade have been cleaned
at April 22nd 2014 (see top panel of Fig. 3 picturing this type
of soiling). The cells shaded by the lichen and algae organisms
are incapable of providing the theoretical mpp-current that
non-shaded cells would be capable of. Consequently, this
current limitation imposed by partly shaded cells causes non-
shaded cells to operate at a lower current but an higher voltage
as compared to mpp. Hence, all unshaded cells cause a voltage
increase. Once cleaned, DC-voltages significantly decreased
whilst PR even more significantly increased (see daily
averages of PR, marked in red dots (un-cleaned) and black
dots (cleaned) in the lower curve depicted in Fig. 4). An ~6%
increase in performance according to our simulation model
that includes all loss mechanisms covered in this paper was
attributed to the cleaning event. The voltage decrease is hardly
visible in Fig. 4, because fluctuating module temperature
superimposes large voltage variations. To this end, Fig. 5
illustrates the voltage decrease, due to the module cleaning
event, by plotting DC-voltage as a function of module
temperature for data shown in Fig. 4. An accurate
measurement of module temperature is essential for this and
many other kinds of analysis, and temperature effects will be
discussed in more detail in Section III.
D. Deviations from STC
Deviations from STC concern reflection, spectral effects,
irradiance intensity dependency and temperate, each affecting
PV efficiency. All these deviations have been, are and will be
matter of R&D in PV performance modelling.
Studies on spectral effects have not yet come to conclusive
results; it is for example still unclear, whether overall a gain or
a loss is concerned for c-Si and spectral effects [4, 5].
Furthermore, reflection aspects may become more complex in
the future, with anti-glare or anti-soiling coatings likely to
have reflection-related impact. In addition, innovative module
designs that employ light-catching ribbons (LCR) and light-
reflecting foils (LRF) are being realized by module
producers [6]. We anticipate such devices to be very difficult
to model, as long as transparency on used models and data that
simulates potential gains has not been validated. Moreover,
irradiance intensity dependent performance of standard c-Si
modules can strongly depend on the particular module (or
solar cell) batches delivered to the site.
For irradiance intensity dependence, Fraunhofer ISE in fact
uses only one single performance curve over irradiance
intensity for standard c-Si to model this loss mechanism. This
curve is based on the annually updated average efficiency of
top-tier module producers of standard c-Si, derived from
measurements at Fraunhofer ISE CalLab PV modules of these
top-tier module producers. To illustrate this matter, irradiance
dependency of efficiency for eight selected module types as
recently published [7] is shown in Fig. 6 (for A1-3
978-1-4799-7944-8/15/$31.00 ©2015 IEEE
representing c-Si supplier 1, c-Si supplier 2, high efficiency c-
Si, for B two CdTe suppliers, for C two CIGS suppliers, and
for D representing a-Si). The median value for crystalline
silicon module efficiencies, measured at Fraunhofer ISE’s
CalLab PV modules, is based on ~100 measurements in 2013
and the gray area indicates the typical range (5th95th
percentile).
When using other irradiance intensity performance curves
for c-Si than the average, as described above, we do so only, if
based on power rating measurements for the very modules
delivered to the specific PV power plant assessed. This way,
we believe performance testing can truly help reducing
uncertainties of otherwise pure predictions, e.g. using PAN
files that not necessarily reflect actual performance of module
batches (and solar cell batches contained in the modules,
respectively) delivered to the specific PV power plant.
Power rating measurements of modules according to IEC
61853-1 may help to reduce uncertainties in the future. This
also holds for temperature losses, where standardization as
well as insights from thermal modelling leads to better defined
module over-temperature coefficients. However, although
these standards give insights when comparing module types of
different design (e.g. standard front-side glazing vs. double
glazing regarding over-temperature), uncertainty of losses in a
PV power plant might be still very high, as long as
environmental factors remain uncertain. Here, we give a brief
example on module temperatures and associated losses on a
system level in Section III.
Fig. 6: Irradiance intensity dependency of various PV module types
as measured by Fraunhofer ISE in year 2013 [7].
E. Actual PV capacity and further DC- losses
Industrial quality assurance services have been well-
established in the past years to address uncertainty involved
regarding actually installed PV capacity. Usually, this
comprises measuring IV-curves of selected strings to evaluate
in how far nameplate actually meets real on-site capacity.
Thermography imaging of modules, combiner boxes and
cabling connections to inverters are usually performed at the
same time, as well as any further due diligence services
requested for by usually the investor.
Relatively high uncertainty remains, however, with the
lower end of 2% uncertainty stated in Table I being fairly
optimistic, and perhaps only possible for small scale systems.
This remaining uncertainty is primarily caused by irradiance
intensity measurement in the plane-of-array, but also module
temperatures play a significant role. For large-scale plants,
often an average temperature needs to be estimated, because
temperature measurements in practice will only be made at
one location, although PV module tables are very large and
temperature spread amongst the modules at different locations
of the table may occur. An example related to module
temperature and thereto associated losses is given in the
following discussion section of this paper.
F. Inverter losses
The potential to reduce uncertainty in this category is
somewhat limited, with inverter exhibiting efficiencies as a
function of power output that can be modelled quite well. This
is actually very fortunate, illustrated in Fig. 7 by a reasonable
efficiency fit (red line) to measured data (colored dots).
Fig. 7: Annual inverter efficiency data (dots) and model (red line).
On the one hand, the fairly high upper end of uncertainty of
category “B: Shading” associated to inter-row shading is
attributed to inverter behavior. Here, this high uncertainty is
currently assumed by us for particular shading cases and
unknown inverter, which may not find the mpp voltage
accurately in case of shaded module substrings, mounted
horizontally, and thereby switching bypass diodes integrated
into the modules due to the shade. For the case of inter-row
shading, a combination of the inverter model as shown in
Fig. 7 and the shading model as shown in Fig. 2 may prove
978-1-4799-7944-8/15/$31.00 ©2015 IEEE
that this is a needless worry, thereby removing the uncertainty
related to energy loss for “inverter behavior under specific
shading conditions” entirely.
G. AC-losses
AC-losses may be relatively low, but surely they need to be
considered, if the point of electricity measurement the yield
study relates to include AC-grid parts. In the future, we may
see very slightly increased losses and uncertainties of those
losses here, as the need of reactive power to be fed-in by PV
power plants is likely to gain further relevance. Much more
significant is the very high uncertainty related to grid-
availability or grid-capability to take up the electricity
(potentially) generated by PV power plants. In fact, we see for
a number of PV systems that are evaluated in terms of
performance, that their grid-connection has not been finished
and the grid in incapable in taking up the entire power the PV
power plant is theoretically capable to produce. However, this
is policy- and grid-related, and from the technical PV-
performance point of view we currently assume these effects
to be methodologically excluded. The PV-performance related
AC-losses contribute insignificantly to overall uncertainty
from the technical viewpoint, and therefore we omit any
further descriptions of ongoing modelling activities in this
category.
III. CASE STUDY ON INVOLVED UNCERTAINTIES
Following the analysis of each loss mechanism, as
described in Section II, measured and expected PV
performance can be compared. To this end, also external
meteorological data is frequently used, which allows to further
validate quantities measured on-site. These two kinds of data,
on-site PV performance and meteorological data, are then fed
into a simulation software at Fraunhofer ISE denoted Zenit.
The software Zenit includes established procedures to
calculate PV energy loss mechanisms as well as
meteorological tools, such as irradiation transposition models,
and PV system related tools, such as PV low-light
performance curves and inverter efficiency models, and is
described in detail elsewhere [8].
In a case study, we used the Zenit software and approaches
detailed in this paper, to perform an analysis of temperature
losses modelled by the simple over-temperature approach, i.e.,
temperature losses simulated directly from given ambient
temperature and GPOA by meteorological data. This we
compared to measured performance data of some ~70 systems
from our in-house monitoring campaign installed in Germany.
Figure 8 illustrates the modelling approach of using only an
over-temperature coefficient in the energy yield prediction.
The model is reflected by the red line in the graph, showing
25K difference in ambient to module temperature at irradiance
intensity in the plane-of-array of 1000 W/m2. Clearly, the
deviations of temperature are very large, but as indicated by
the color bar, this is mostly due to wind-speed.
In order to investigate, in how far one can match the
measured temperature losses of systems by this simplistic
model, we fitted on an annual basis data from a number of our
monitoring installations, distributed all over Germany, using
this linear temperature model. For all data it has been assured
that in no case sensor-detachment of the temperature
measurement device (attached to the backside of the modules)
has occurred.
Fig. 8: Over-temperature ΔT measured during one year excluding in
one of the systems evaluated in the study presented here (dots) and
simplistic over-temperature model (red line), see text.
Fig. 9: Normal distribution of fitted over-temperature coefficients
“k_mod” that have been obtained by linear regression of monitored
temperature losses in ~70 German PV installations including ~400
years of data.
Fitting the data using the linear temperature model for in
total ~400 monitoring-years resulted in the same number of
~400 k_mod values, which exhibited reasonable symmetrical
978-1-4799-7944-8/15/$31.00 ©2015 IEEE
normal distributions for each system type “ground mounted”,
“mounted on large-area industrial roofs” and “roof-
integrated”. The normal distribution fit of each set of k_mod
for these system types is shown in Fig. 9, with colored areas
indicating one standard deviation. As can be seen in these
distributions, over-temperatures assumed in simulation (stated
in the graph for each system type) correspond quite well with
monitored data for ground-mounted and even roof-integrated
systems, given the limitations of the model itself. For large-
area rooftop installations, however, it is clear that observed
temperature loss from monitoring data is not reflected by a
yield prediction model that uses 30 K for the k_mod parameter
and that system type. Here, temperature losses in the
simulation would be systematically higher than those actually
occurring in PV system operation for such systems.
IV. CONCLUSIONS
We presented a summary of uncertainty sources related to
various loss mechanisms differentiated in the energy loss
chain of Fraunhofer ISE yield prediction reports.
The overall initial uncertainties of yield prediction reports
can be very high, up to ~11%. Fraunhofer ISE states in its
yield prediction reports our current best-estimate of
uncertainty for each modelling step. We would like to
encourage everyone preparing yield prediction reports to do
the same, especially when the report is prepared within a
commercial setting.
Uncertainties of energy yield prediction reports may get
reduced significantly following the first year of PV system
operation. This is only possible, however, if the quality of on-
site monitoring systems are reasonably met. First and
foremost, irradiation data needs to be of high quality. This
then allows for state-of-the site-adaptation processes [2] to
significantly reduce uncertainty of the irradiation category
down to +/- 2.5% for global horizontal and +/- 4% for direct
normal irradiation. Fraunhofer ISE therefore regularly
performs verification measurements on-site using own
equipment, verifying that sensors and data acquisition used in
the plant record data that is well within confidence levels
needed.
Finally, it has to be noted to that in our daily practice we
frequently find quality-limitations of operational monitoring
of PV power plants to be in many cases unreliable and
inaccurate. Even if the PR is one of the most important values
for financing institutions, no standard is currently available
which guarantees data acquisition is reliable and has a defined
quality, allowing for appropriate PV performance evaluation.
Here, on-site performance verification allows the validation of
operational monitoring systems integrated into PV power
plants.
REFERENCES
[1] N. H. Reich, B. Müller, A. Armbruster, v. W. G. J. H. M. Sark,
K. Kiefer, and C. Reise, "Performance ratio revisited: is PR > 90%
realistic?," Progress in Photovoltaics: Research and Applications,
vol. 20, pp. 717-726, Sep 2012
[2] Site-adaptation techniques offered as a commercial service; see
http://geomodelsolar.eu/services/site-adaptation (June 14th 2015)
[3] B. Müller, T. Reis, A. Driesse, C. Reise, Maximizing the Yield
of Large PV Power Plants: What Can We Learn from
Monitoring and Simulation? Proc. of the 27th European PV Solar
Energy Conference and Exhibition, Frankfurt, pp. 3775 - 3781
[4] D. Dirnberger, G. Blackburn, B. Müller and C. Reise, "On the
impact of solar spectral irradiance on the yield of different PV
technologies," Solar Energy Materials and Solar Cells, vol. 132
pp. 431-442, 2015.
[5] D. Dirnberger, B. Müller and C. Reise, "On the uncertainty of
energetic impact on the yield of different PV technologies due to
varying spectral irradiance," Solar Energy, vol. 111, pp. 82-96,
2015.
[6] Dirnberger, D., B. Müller, et al. (2015). "PV module energy
rating: opportunities and limitations." Progress in Photovoltaics:
Research and Applications, published online
[7] S. Zhang et.al, 335Watt World Record P-type Mono-crystalline
Module With 20.6 % Efficiency PERC Solar Cells, this
conference
[8] Müller, B., L. Hardt, et al. (2015). "Yield predictions for
photovoltaic power plants: empirical validation, recent advances
and remaining uncertainties." Progress in Photovoltaics:
Research and Applications: published online
978-1-4799-7944-8/15/$31.00 ©2015 IEEE
... A reduction in the uncertainties can lead to a higher value of energy yield for a given exceedance probability and hence a stronger business case. Reich et al. [18] estimated the combined overall uncertainty of the energy yield to fall in a range between 5% and 11%; in the study, the uncertainty on various effects such as irradiation, shading, soiling and inverter losses was taken into account. In another study, Müller et al. [19] have calculated the variation of the overall uncertainty of the energy yield over the lifetime of a PV plant and compared the findings with data from a portfolio of 26 systems located in Germany and Spain. ...
... This is of particular importance for systems where a guaranteed yield is requested together with a PR. Reich et al. [18] give an overall uncertainty on the energy yield because of various effects in the range of AE5-11%. Richter et al. [21] give a value of AE6-8% for the energy yield and AE2-6% in PR. ...
Article
Full-text available
In photovoltaic (PV) projects, it is important to establish a common practice for professional risk assessment, which serves to reduce the risks associated with related investments. The objective of this paper is to present a methodology on how to improve the current understanding of several key aspects of technical risk management during the PV project lifecycle, with the identification of technical risks and their economic impact. To achieve this, available statistical data of failures during a PV project have been collected with the aim to (i) suggest a guideline for the categorisation of failure and (ii) develop a methodology for the assessment of the economic impact of failures occurring during operation but which might have originated in previous phases. The risk analysis has the aim to assess the economic impact of technical risks and how this can influence various business models and the levelised cost of electricity. This paper presents the first attempt to implement cost-based failure modes and effects analysis to the PV sector and to define a methodology for the estimation of economic losses because of planning failures, system downtime and substitution/repair of components. The methodology is based on statistical analysis and can be applied to a single PV plant or to a large portfolio of PV plants in the same market segment. The quality of the analysis depends on the amount of failure data available and on the assumptions taken for the calculation of a cost priority number. The overall results can be linked to the cost of periodic and corrective maintenance and form the basis to estimate the impact of various risk and mitigation scenarios in PV business models. Copyright
... DISCUSSION AND RELEVANCE Whilst the highest efficiencies for small PV cells in laboratory contexts are near 32%, commercial PV modules have a maximum rated efficiency of close to 22%. In operational PV installations these efficiencies decline further to be in the range of 13 to 17%, because a PV system [3][4][5] is (i) exposed to solar irradiation intensity, collimation and spectra that vary but are different from those under which standard rating procedures are conducted, (ii) subject to soiling, (iii) has a lower than expected inverter efficiency, (iv) inefficient maximum power point tracking and (v) suffers thermal de-rating of PV modules (e.g. 0.4% loss per °C for silicon PV modules). ...
... In particular, the following possible additional scenarios are considered for each input parameter: -insolation variability: 5 years of data; 22 years of data but from a location characterized by a high variability; -solar resource: ground measurements; site adaptation technique (combination of long-term satellite data and short-term series of measured data [9]); -transposition model: no transposition model (GTI data are available); transposition model proven to be the worst for a specific location [10]; -ambient temperature variability: highest value in [1] -temperature effect: using a temperature model that does not account for wind effect (e.g. NOCT formula), proven to be bad for Bolzano [6]; -performance loss rate: highest value in [1]; -soiling loss: highest value in [11]; -shading loss: 5% relative uncertainty (k=1); -spectral mismatch loss: 2% relative uncertainty (k=1); -nominal power of the modules: highest value in [1]; -inverter efficiency effect: highest value in [1]. ...
Conference Paper
Full-text available
In PV projects, it is important to understand how the variability and associated uncertainties of technical risks occurring during the planning phase are calculated and how the values are distributed in terms of probability. These aspects are essential for the calculation of the exceedance probability of the expected energy yield and how this is influenced by the overall uncertainty. A reduction in the uncertainties can lead to a higher values of energy yield for a given exceedance probability and hence a stronger business case. In this paper, the combined effect of uncertainties of various technical risks is calculated and discussed for several uncertainty scenarios and the impact is calculated in terms of exceedance probability (e.g. P90/P50 ratio).
Conference Paper
Full-text available
Yield predictions are performed to predict the solar resource, the performance and the energy production over the expected life time of a Photovoltaic (PV) system. In this study we compare yield predictions and monitored data for 26 PV power plants. The monitoring data includes in-plane irradiance for comparison with the estimated solar resource and energy yield for comparison with predicted performance. The results show that, due to increased irradiance in recent years (" global brightening ") the yield predictions systematically underestimate the energy yield of PV systems by about 5%. As common irradiance databases and averaging times were used for the yield predictions analyzed in this paper, it is concluded, that yield predictions for areas where the global brightening effect occurred in general underestimated the energy yield by the same magnitude. Using recent satellite derived irradiance time series avoids this underestimation. The observed Performance Ratio of the analyzed systems decreases by 0.5 %/year in average with a relatively high spread between individual systems. This decrease is a main factor for the combined uncertainty of yield predictions. It is attributed to non-reversible degradation of PV cells or modules and to reversible effects, like soiling. Based on the results of the validation the combined uncertainty of state of the art yield predictions using recent solar irradiance data is estimated to about 8%. Keywords: PV systems, yield prediction, performance ratio, solar resource assessment, global dimming and brightening
Article
Full-text available
Yield predictions are performed to predict the solar resource, the performance and the energy production over the expected lifetime of a photovoltaic (PV) system. In this study, we compare yield predictions and monitored data for 26 PV power plants located in southern Germany and Spain. The monitoring data include in-plane irradiance for comparison with the estimated solar resource and energy yield for comparison with predicted performance. The results show that because of increased irradiance in recent years (‘global brightening’) the yield predictions systematically underestimate the energy yield of PV systems by about 5%. Because common irradiance databases and averaging times were used for the yield predictions analysed in this paper, it is concluded that yield predictions for areas where the global brightening effect occurred in general underestimated the energy yield by the same magnitude. Using recent satellite-derived irradiance time series avoids this underestimation. The observed performance ratio of the analysed systems decreases by 0.5%/year in average with a relatively high spread between individual systems. This decrease is a main factor for the combined uncertainty of yield predictions. It is attributed to non-reversible degradation of PV cells or modules and to reversible effects, like soiling. Based on the results of the validation, the combined uncertainty of state of the art yield predictions using recent solar irradiance data is estimated to about 8%.
Conference Paper
Full-text available
Today’s PV systems tend to get larger, so even small improvements in Performance Ratio (PR) may lead to remarkable financial benefits. The PR depends on several design decisions, and in most cases, the advantage of a design decision may be confirmed in advance by simulation only. To validate system simulation models for this purpose, we compare high resolution system performance data to the results of our yield prediction models. Even if several technical losses of a PV system cannot be measured directly in operating commercial power plants, the individual loss models may be validated by looking at certain irradiance ranges or certain geometric conditions. Assessment of inverter power limitation will serve as one example of our approach. The full advantage of a detailed analysis of monitoring data is then shown for the effect of row-to-row shading. While the uncertainty of PR values predicted by time step simulations may be relatively high for single time steps and may be biased for annual values, it is still possible to get valid results for system optimizations.
Article
Full-text available
In this study, we investigate the performance ratio (PR) of about 100 German photovoltaic system installations. Monitored PR is found to be systematically lower by ~2–4% when calculated with irradiation data obtained by pyranometers (henceforth denoted as PRPyr) as compared with irradiation amounts measured by reference cells (denoted as PRSi). Annual PRSi for the ~100 systems is found to be between ~70% and ~90% for the year 2010, with a median PR of ~84%. Next, simulations were performed to determine loss mechanisms of the top 10 performing systems, revealing a number of these loss mechanisms may still allow for some optimization. Despite the fact that we do not see such values from our monitoring data base up to now, we believe PRSi values above 90% are realistic even today, using today’s commercially available components, and should be expected more frequently in the future. This contribution may help in deepening our knowledge on both energy loss mechanisms and efficiency limits on the system level and standardization processes of system-related aspects.
Article
This article sheds new light on photovoltaic (PV) module rating according to predicted yield rather than power measured at standard testing conditions (STC). We calculate module performance ratios (MPR) for measured characteristics of eight different module types and compare them with a reference MPR calculated with typical crystalline silicon characteristics. In place of the not yet existing standardized weather data, we use commercially available weather data for three different locations. The reference MPR for the three locations were 95.5%, 94.6%, and 91.0%, respectively, with differences to the other module types of ±8% at maximum. MPR was calculated with reference to nominal power, and—following IEC 61853—without consideration of potential degradation. The strongest contribution to the initial differences between the module types was due to differences in irradiance dependency. Standard uncertainties for all initial MPR values were calculated and range from 1.8% to 3.0%, including STC power uncertainty. We propose a module rating method that indicates whether a module type's performance is significantly above, below, or essentially equal to the reference. The method evaluates the MPR difference between module type and reference, taking uncertainty into account. Significant differences were only found between modules with obviously different characteristics, but not between the crystalline silicon module types under scrutiny. As the uncertainty analysis did not cover degradation and influences due to the use of not standardized weather data, a sensitivity analysis was performed. Long-term degradation can change the comparative energy rating significantly, whereas the selection of tilt angle and assumptions regarding module operating temperature did not have a strong effect. Copyright © 2015 John Wiley & Sons, Ltd.
Article
Abstract Varying spectral irradiance affects performance of photovoltaic (PV) modules depending on their band gap and spectral response. Therefore, energy rating or yield prediction procedures should take into account the irradiance that is effectively available for a specific PV technology. This effective irradiance differs from broadband solar resource information first due to the different spectral sensitivity of PV devices and pyranometers, and second due to the difference between real and reference spectral distribution. The combined difference, also referred to as ‘spectral impact’, is to date considered in form of fixed spectral gain or loss factors applied to the broadband solar resource – if it is considered at all. Values for the monthly or annual spectral impact have been reported for several locations; however without estimates of their uncertainty. This article addresses the question of how large the uncertainty of the spectral impact is. Contributions to the uncertainty of the spectral impact are discussed, with focus on the method of determining the spectral impact by means of measured spectral irradiance and calculated spectral mismatch factor (MM). Using a simplified procedure, the uncertainty of spectral impact was estimated from the average spectral mismatch uncertainty. It was found to be of the same magnitude as the spectral impact itself, and dependent on the PV technology. For five single-junction PV technologies, amorphous silicon, cadmium telluride, standard crystalline silicon, high-efficiency crystalline silicon and a wide-band-gap CIGS technology, the estimated uncertainty values were 1.8%, 1.4%, 0.9%, 1.0% and 1.2%, respectively. The results were obtained from a specific measurement campaign conducted from 01.06.2010 to 31.12.2013 in Freiburg im Breisgau, Germany, but the order of magnitude is considered to be typical. In order to validate the simplified uncertainty estimation method, a sensitivity analysis was carried out to evaluate the influences of calibration and drift of the spectroradiometer, high-angle-of-incidence conditions, the available wavelength range of the spectroradiometer and results obtained with different spectroradiometers. With the smaller wavelength range of 350–1050 nm, which is a typical range for various instruments used in many scientific publications, the calculated spectral impact was significantly smaller. The other influences were found to affect the result only within the estimated uncertainty limits. The uncertainty values presented here should be considered lower limits until they are backed up by further analysis.
Article
This article presents results from investigating the impact of varying spectral irradiance on the performance of different PV technologies. Relative gains or losses were quantified for five typical PV technologies with different band gaps and, consequently, different spectral responses using spectral irradiance measured from 01.06.2010 to 31.12.2013 in Freiburg im Breisgau, Germany. With the spectrally resolved data, the effectively available annual irradiation per technology was calculated using the spectral mismatch factor, and compared to the total broadband irradiation measured by a pyranometer. The process used to calculate the spectral impact produces a result that can directly be used to estimate the effectively available irradiation for yield prediction or energy rating. The annual spectral impact was +3.4% for amorphous silicon, +2.4% for cadmium telluride, +1.4% for crystalline silicon, +1.1% for high-efficiency crystalline silicon and +0.6% for small-band-gap CIGS. Technologies with a large band gap exhibited spectral gains in summer and spectral losses in winter, and vice versa for small-band-gap technologies. The results are discussed and interpreted with consideration to uncertainties and results published so far. Furthermore, it was investigated in how far average photon energy (APE) can be used as a quantitative indicator for the spectral impact. In summary, it was found that using APE does not present actual advantages over using the spectral mismatch factor, and should rather be used for qualitative than for quantitative evaluations.