PV System Performance Assessment

By James Mokri, SJSU, and Joe Cunningham, CentroSolar

In collaboration with the SunSpec Performance Committee

Practices, Methods and Guidelines to Assess Performance of

Existing Systems

Version 2.0 June 2014

Published by the SunSpec Alliance
www.sunspec.org

1
Metrics and methods to assess performance
of existing systems to aid bankability of
PV asset class
Determining and evaluating system performance based on
actual weather and actual system characteristics is
critical to developing creditability for PV as an asset
class.
takeholders of existing photovoltaic (PV) ROI due to significantly lower annual energy

S solar energy systems are typically interested

in system performance for operation and
maintenance planning, commissioning, performance
production.
Bankability of PV assets requires that investors
understand the reliability of modeling and actual
guarantees and for making investment decisions. performance data in support of their investment
Monitoring companies are developing data analysis decisions and how it is related to:
methods to process real-time data for their specific
systems and performance metrics. However, a  Equipment
literature review of metrics in common use by  Location
companies found that various analytical methods
are used to calculate the same metric, or they are  Design
using one analytical method with varied results due  Contractor and Installation Technique
to the environment of the system. Both are
problematical because they result in different  Maintenance
interpretations It would be desirable for stakeholders to have
consistent definitions, methods, and agreement
For example, the commonly used metric of
regarding the objective of the metric. This would
Performance Ratio (PR), as defined by IEC61724
enable better classification of the performance of
and NREL, may be appropriate for annual
solar assets across technologies and location.
comparison of systems with the same climates but is
Consistent performance standards would also help
not appropriate for shorter term or system
streamline the bankability assessment for solar
comparisons in differing climates. Specifically, if
assets.
PR is used to evaluate a system in San Francisco,
CA, compared to a similar system in Daggett, CA, This article identifies representative metrics in
incorrect conclusions would be reached. Using current use, summarizes the method and level of
PVWATTS to represent an actual system, a 100kW effort to calculate the metrics, reviews the objective
system in San Francisco with latitude tilt has a of the metrics, estimates the metric uncertainty
calculated PR of 0.73 with an output of 145,000 level, and recommends which metric is appropriate
kWh/year, while a 100kW system in Daggett with for which purpose/objective.
latitude tilt has a PR of 0.69 with an output of The following four performance metrics are the
171,000 kWh/year. Even with a lower PR, the focus of this article:
Daggett system has higher output and therefore
higher performance.  Power Performance Index (PPI) of actual
instantaneous kW AC power output divided by
If PR is used to make an investment decision in one expected instantaneous kW AC power output.
of these systems, all other factors being equal, the
investor would choose San Francisco with a lower
2
  Performance Ratio with temperature Calculations were performed to evaluate the
corrected final yield using weighted-average uncertainty range for various metrics. Data was
cell temperature (CPR). Note that Performance obtained from exiting systems which had weather
Ratio is commonly defined without temperature stations and had accessible data through on-line
correction. monitoring sites.
 Energy Performance Index (EPI-SAM) of Performance Assessment
actual kWh AC energy divided by expected Objectives
kWh AC energy as determined from an
accepted PV model, such as SAM, using actual The objectives for performance assessment can best
climate data and assumed derate factors. be summarized from an owner’s perspective by the
questions that are often asked:
 Energy Performance Index (EPI-
REGRESSION) of actual kWh AC energy • How is my system, or a portion of my
divided by expected kWh AC energy as system, performing currently in comparison
determined from a polynomial regression to how I expect it to perform at this point in
equation having coefficients determined from its life?
actual operating and climate data collected • How is my system performing for both the
during the model “training” period. short-term and long-term in comparison to
Some conclusions of this study show how the above how it is capable of performing with its
four metrics are applicable for the following given design, site location and baseline
performance assessment objectives: performance?
 Monitoring of a specific PV system to • How is my system performing over an
identify degraded performance and need for assessment period in comparison to other,
condition based maintenance. similar systems in similar climates?
Recommendations, including varied levels of • How is my system performing compared to
uncertainty, are to use EPI-SAM or EPI- the last assessment periods? This trending
Regression or CPR. model is useful for maintenance objectives.
 Commissioning of a new system, re- • How can I develop metrics in support of
commissioning, or assessment after major accurate prediction of future energy yield
maintenance and to set a baseline for future and ROI for reliable investment assessment.
performance measurements and comparisons.
Recommendation is to use PPI and EPI metrics. • During commissioning, what metrics should
be used to set a baseline for future
 Determination of specific industry performance assessments?
parameters, such as Yield or Performance Ratio,
to allow comparison of systems in different One objective of a performance assessment is to
geographic locations for design validation or detect changes in system performance; usually
investment decisions. Recommendation is to decreases in performance, to allow the system
use Yield, PR, CPR and/or EPI depending on owner to investigate and potentially perform cost
the level of effort and level of uncertainty. In effective maintenance. This can be done best on a
some cases, depending on the objective, relative scale where the specific performance of the
combinations of these metrics are most useful. system is compared to itself which reduces adverse
Although this study was intended for metrics that effects of modeling input assumptions and
apply to fixed flat panel PV module technology uncertainty.
used on systems of greater than 100kW DC, the
Another objective is to determine if a new system,
metrics are actually helpful for any fixed flat plate
or an existing system having completed major
panel PV system size. Further explanations are
maintenance, has instantaneous power output and a
shown on the application map of Figure 1.2.
0 to 6 month energy output consistent with
3
predictions by the design model. This is also
considered a commissioning activity and since there
is no long-term operating data, the results are
directly dependent on the validity of the model and
input assumptions which both increase uncertainty.

It should be noted that system performance is

different than system value or system reliability.
The performance of a system is indicated by the
actual AC energy or power output relative to its as-
designed or as-built capability. Deviations from
100% can be caused by many factors, including
errors or incorrect assumptions during design, poor
installation workmanship, equipment failure or
degradation, etc. The value of a system is related to
the system lifetime cost relative to the AC energy
output, often referred to levelized-cost-of-energy
(LCOE). Also, performance is different than
reliability although performance is dependent upon
reliability.

Figure 1.2 shows the relative types of assessment

and the applications.

4
 Fig. 1.2 - Performance Assessment Map showing
applicability of recommendations covered by this report

Cost Effective
Approaches to
Proprietary Proprietary Proprietary Performance
Maximum

2% to 5%

Algorithm Algorithm Algorithm Assessment

PPI, PR with PPI, PR with PPI, PR with

10% to 20% 5% to 10%
Moderate

temperature temperature temperature

compensation compensation compensation
factor, EPI factor, EPI factor, EPI
Level of Effort:

Uncertainty:

Inverter kWh meter, Inverter kWh meter, Inverter kWh meter,

Minimum

Utility billing, PPI, Utility billing, PPI, Utility billing, PPI, SunSpec
PR w/o adjustments PR w/o adjustments PR w/o adjustments Performance
Assessment
Focus

System Size: Small Medium Large

<20kW >100kW >10 MW
Asset Class: Residential Commercial Large Commercial,
PPAs

The recommendations were developed to be applicable to fixed flat panel PV module technology.

5
Current Industry this metric is Performance Ratio and it is
used regularly to compare systems.
Performance Metrics – However, it may result in incorrect
Literature Survey conclusions if the systems being
The review of currently used performance compared are in different locations with
metrics included information from NREL, different irradiance and temperature.
Sandia, IEC, equipment suppliers, and other
Performance metrics can first be divided
organizations. Some metrics appropriately
into instantaneous, short-term, and long-
use a ratio of actual performance divided by
term assessment periods. Various
expected performance, called Performance
degradation mechanisms and intermittent
Index (PI). Some methods have established
anomalies develop and occur over long-term
acceptance criteria which define the
periods so both periods are needed to
minimum output and are used primarily
complete an assessment. Instantaneous
during commissioning. Inputs used in
output is based on power and is denoted by
calculating expected performance included
kW (power). A long-term assessment
as-build system component ratings and
period, such as weekly, monthly, or annually
technology, irradiance, ambient temperature,
is based on energy and yield, and is denoted
wind, mounting, module temperature, and
by kWh (energy).
typical condition dependent derate factors.
Performance metrics can also be divided
The condition dependent derate factors are
into absolute and relative values. An
difficult to determine and they have a large
absolute value can be used to evaluate a
influence on the performance calculation,
system by comparing it to industry-wide
and also introduce significant uncertainty
values resulting in a figure of merit for the
into the calculations.
system. A relative performance metric can
In principle, performance assessment could be used to trend a specific system using
be based on any of the following: trend plots of the metric and associated
parameters. Both the absolute and relative
 Actual output divided by actual solar
metrics provide input to troubleshooting of
input. This metric is representative of
degraded systems. Measurement
overall system efficiency and a normal
uncertainty and error analysis should be
system would have a value on the order
used to define a tolerance band to avoid
of 0.1, largely dependent on the module
reaching inappropriate conclusions.
efficiency. No analytical PV model is
needed in this case. This metric has Some metrics, such as Yield and
limited use most likely due to the Performance Ratio are independent of a PV
negative perception of a low value model, whereas Performance Index is the
around 0.1. actual performance divided by the calculated
 Actual output divided by expected expected performance and is therefore
output. This metric is largely dependent dependent upon an accurate PV model.
on the system design, quality of Initial review of industry practice found
installation, and the accuracy of the PV various performance metrics as shown in
model. A normal system would be on Table 2.1 in the Appendix.
the order of 1.0. This metric is used and
can be based on either power or energy.
 Actual output normalized divided by
actual input normalized. An example of
6
Yield Summary of Effective
The standard Yield metric is considered to Performance Metrics
be the “bottom-line” indication of how well The industry has used various metrics, often
a system is performing since the purpose of with similar names but different calculation
the system is to maximize energy output for methods, or with different names and similar
a given system size; however, it does not calculation methods. Some metrics and
account for weather conditions or design and calculations presented in technical papers
can only be applied for a consistent are not effective for the purpose intended.
assessment period (such as annually). Since As the industry has evolved, data has
Yield increases proportionally with hours of become more available, and analyses easier
operation, insolation, and lower temperature, to perform; newer methods have been
a high yield due to unusually high insolation proposed and used. Based on evaluation of
can be misleading and potentially even mask these various metrics, those that are
a case of a degrading system. Conversely, a considered appropriate for assessments are
system with an unusually low insolation summarized in Table 2.2.
may be incorrectly judged to have poor
performance. If systems are being In general, performance assessment is the
compared using Yield, the hours of process of measuring or monitoring actual
operation, insolation, and cell temperature performance and comparing it to expected
should be equivalent for a fair comparison. performance.
The basic Yield equation is shown below as Either the actual performance or the
equation 1: expected performance must be adjusted to
account for the actual weather and derate
factor conditions. One approach is to adjust
the actual system kW AC output “up” to
STC (e.g. apply a ratio of 1000 W/m2 /
The value of a system ultimately comes Gactual) and compare this to the expected
down to annual AC energy output relative to STC system output from PV model
system cost. Therefore, Yield is a measure calculations.The other approach is to adjust
of system value rather than performance. the STC output from PV model calculations
Performance Index “down” to the actual condition (e.g. apply a
ratio of Gactual / 1000 W/m2). The second
Performance Index (PI) as typically used by approach is appropriate and more commonly
the industry represents the ratio of actual used by the industry.
output (either power or energy) of a system
divided by the expected output. The Performance Index (PI) is typically the
expected output was calculated using an direct ratio of actual output divided by
accepted PV model, such as the NREL expected output, and is obviously different
System Advisor Model (SAM), or a than a ratio of output divided by input such
regression model, therefore, the accuracy as is used in an efficiency equation.
and uncertainty of the PI value is dependent Performance Ratio (PR), as defined by
on the accuracy and uncertainty of the NREL and IEC, is a normalized version of
model. output divided by input so its value is not
similar to a system efficiency of around 10%
but rather is around 70%. The normalizing

7
approach of including the DC STC rating
and irradiation ratio, effectively converts the
PR to a ratio of actual output divided by a
“rough estimate” of expected output. If
compensation factors in addition to actual
irradiation are added to PR, such as
temperature, balance of system losses, etc.,
it converts PR to a ratio with expected value
in the denominator and is then similar to PI.
The simple algebra is shown later.
Energy Performance Index (EPI) is a ratio of
actual kWh AC divided by expected kWh
AC using actual climate data over the
assessment period as input to an accepted
PV system model, such as SAM with all
relevant derate parameters included, or as
input to a “trained” regression model.
“Trained” refers to the process of using
actual system historical data to solve for
regression equation coefficients. Therefore,
EPI (either SAM or Regression methods)
incorporates the most complete metric for
performance assessment.
In the paragraphs that follow, the four
metrics which are considered to be
appropriate for performance assessment are
discussed.

8
 Table 2.2- Summary of Performance Metrics

METRIC PURPOSE METHOD UNCERTAINTY

PR-Performance Ratio Maintenance (kWh/Rated kWDC) / (kWh/1000) High - 15% to 20%
CPR – Temperature Maintenance [kWh/(Rated kWDC*Temp Corr. )] Moderate - 10% to 15%
Corrected PR / [kWhsun/1000]
EPI – Energy Perf. Index Maintenance, Actual kWhAC / Calc. SAM Moderate - 10% to 15% (model
SAM model Commissioning, kWhAC dependent)
Financial
EPI – Energy Perf. Index Maintenance & Actual kWhAC / Calc. Regression Low - 5% to 10%
Regression model Commissioning, kWhAC
Financial
kWh Production Maintenance Compare AC kWh Period to High - 15% to 20%
Period
Yield Financial only kWh per DC Watt Low - 5% to 10%
PPI – Power Performance Commissioning & Measured kW power Output vs. Low to Moderate - 5% to 15%
Index Troubleshooting Calc. kW power Expected (model and measurement
dependent)

Acronyms:
PI = Performance Index, ratio of actual divided by expected
PPI = Power Performance Index, instantaneous actual power divided by expected power
PR = Performance Ratio
CPR = Temperature compensated Performance Ratio
EPI = Energy Performance Index
kWhAC = AC Energy at system output at utility meter
kWDC = DC rating of array at standard test conditions (STC)
SAM = System Advisor Model, from NREL
KTemp = Temperature compensation factor based on (TCell-TSTC)
kWhSun = Total in-plane solar irradiance
Performance Ratio which could potentially be misinterpreted as
a degrading system. Hourly data also has
Performance Ratio (PR), as defined by variation from morning to afternoon that is
IEC61724 and NREL, is a metric commonly difficult to interpret.
used, however one shortcoming in the basic
PR is that normal temperature variation The seasonal variation of PR can be
influences PR and is not included in the illustrated using PVWATTS to represent an
basic equation. Specifically, cases with low actual system to calculate monthly AC kWh
temperature and moderate irradiation (such and monthly irradiation. A 100kW system
as late winter) result in higher PR and cases with latitude tilt in Sacramento was
with high temperature and moderate arbitrarily selected and analyzed resulting in
irradiation (such as late summer) will result the plot shown in Figure 2.1. It would
in lower PR. A normally operating system appear that the system performance was
typically has a declining PR in the spring, degrading February through July.

9
Figure 2.1 – Basic PR Seasonal Variation Without Temperature Correction

Also as discussed above, PR is more appropriate to trend a specific system or to compare

systems in similar geographic locations. If PR is used to evaluate a system in San Francisco,
CA, compared to a similar system in Daggett, CA, incorrect conclusions would be reached.
Even with a lower PR, the Daggett system has higher output and therefore higher performance.
One of the advantages of using PR is that the expected performance is not calculated, therefore, a
PV computer model is not needed and the inaccuracies and uncertainty introduced by the model
and the derate-factor assumptions are avoided.
Long-Term assessment is needed to identify system degradation due to intermittent faults, out-
of-service time (outages), unavailability, low light performance, angle of incidence effects, solar
spectrum effects, light or potential induced degradation, and other conditions that cannot be
detected during the Short Term assessment period using methods such as those used for
commissioning..
The basic PR calculation uses the standard yield equation in the numerator and the actual
measured plane of array (POA) irradiation summed over the assessment period divided by
standard irradiation in the denominator. The units work out to be hours divided by hours. The
numerator is equivalent to the number of hours the system operated at the DC STC rating and the
denominator is equivalent to the number of peak sunhours of irradiation. Both the measured
irradiation and standard irradiance are in terms of meter2, and cancel directly.

Both the numerator and denominator are summations of the measured increment data over the
assessment period. The assessment period can be daily, weekly, monthly, annually. Calculation
of hourly PR is a problem since some hours of the day with zero irradiance result in division by
zero and is undefined. Since hourly data is commonly available, hourly PR was calculated and
plotted for interest.

10
Analysis of hourly data required filtering to eliminate hours with zero irradiance. The Excel
filter function was used in various scenarios such as to include mid-day hours and for irradiance
greater than a defined value, such as 600 kWh/m2. Effectively, this was a “mid-day flash test”.
Filtering levels raise questions and doubts about the calculated PR value; therefore it is preferred
to calculate daily or longer periods. The Excel function of SUMIFS is useful to calculate the
total values for the period, and AVERAGEIFS is useful to calculate average values such as daily
temperature if temperature correction is being used in CPR.

Instructions for Calculating Long-Term Performance Ratio

PR = (kWhAC/DCRated)/(kWhSun/1kW)
1. Install Plane of Array (POA) irradiance datalogger, or obtain access to existing POA
data, or use data from another local site adjusted from horizontal to POA using NREL
DISC Excel spreadsheet and an anisotropic sky model such as the Perez or similar model.
2. Read inverter kWh total on inverter display at beginning of assessment period, or obtain
access to existing monitoring data.
3. Read totals for irradiation from datalogger and kWh from inverter (or from monitored
data) at end of assessment period; calculate differences to obtain actual kWh of irradiance
and kWh of AC energy over the assessment period. For simpler approach for annual PR
estimate, use PVWATTS total annual POA irradiation value. Annual PVWATTS
irradiation is typically less discrepant from actual than monthly PVWATTS POA
irradiation values, however if the weather during the assessment year is different than the
typical year, uncertainty is increased.
4. Calculate Performance Ratio (PR). Calculate the hourly PR using the IEC61724 formula,
Equation 2 above.
5. Compare PR value to typical industry values, or to similar systems in other locations, or
to previous PR values of the same system to establish trend of performance depending on
the purpose of the assessment.
6. Evaluate PR. If PR ± uncertainty is within Long-Term criteria, system performance is
acceptable. Otherwise proceed to investigate performance shortfall of individual
components.

Performance Ratio, Compensated

The basic Performance Ratio (PR) is directly influenced by energy (kWh) output, which is
directly influenced by irradiation (kWh/m2) and inversely influenced by module temperature.
Since the basic PR equation accounts for irradiation, changes in irradiation will have little direct
effect on PR, however, since changes in temperature are not accounted for, the basic PR will
decrease as temperature increases.
In order to use a metric which is more indicative of system condition rather than design or
environmental conditions that are outside the control of the owner, compensation factors can be
added to the basic PR equation. One method to include temperature compensation is to adjust
the DC rating in the numerator using the power temperature coefficient provided on the module
manufacturer’s data sheet relative to the STC temperature of 25°C. Other methods used for

11
hourly calculations weight the compensation factor by the irradiance or energy output for the
hour, or to use factors based on average annual ambient temperature.
Other factors besides temperature also affect PR and are also outside the control of the owner,
such as design, shading, degradation, balance of system, and could be included as compensation
factors; however the basis for estimating these factors to compensate PR is impractical.
Therefore, if compensation other than temperature is desired, it is more practical to calculate
Long-Term Energy Performance Index (EPI) using actual irradiation and temperature in one of
the accepted models, such as SAM or regression model.
If the purpose of the assessment is only to evaluate a specific system, trend analysis using a
temperature compensated PR is reasonable because it is not influenced by the accuracy and/or
uncertainty of a PV model.
Compensation for factors such as cell temperature, KTemp, can be applied to the basic PR to
adjust the DC power rating from Standard Test Conditions (STC), however since temperature
varies continuously with irradiance and weather, an averaging technique must be performed at
each time increment (such as, hourly) and used to calculate a daily average temperature.

Typical hourly data includes night hours when the energy production and irradiance are zero.
Dividing by zero is undefined; therefore, Daily PR should be calculated using the SUMIF
function in Excel to sum the hourly values to obtain the daily sum of kWhAC and kWhSun. A
daily PR would then be obtained using equation (3). Hourly PR values vary from zero to a
maximum either before or after noon depending on conditions and are considered to be of little
use for performance assessment. Averaging hourly PR to obtain daily PR was tried and not
recommended versus summation of the hourly kWhAC and kWhSun values for the day.
Because irradiance and temperature change continuously, it would be beneficial to use a time
increment less than an hour, however for practicality an average hourly temperature is
considered acceptable unless the assessment is for a large critical system. The 2004 King paper,
suggests that hourly averages is acceptable for most assessments, although other experts say
hourly average under-predicts performance due to the thermal lag when irradiance increases.
If additional compensation factors are of interest to be included, such as balance of system
losses, angle of incidence, soiling, shading, long-term degradation, etc, it is more practical to
include them in the Energy Performance Index (EPI) using an accepted PV model, such as SAM,
to incorporate the compensation factors rather than complicating PR.

12
Figure 2.2 –PR Without and With Temperature Compensation

13
Energy Performance Index (EPI) – SAM or equal:
When compensation factors are added to the PR equation, the equation is equivalent to
Performance Index of actual energy divided by expected energy for the assessment period.
Note that the PR equation which includes compensation for temperature or other factors is
identical to the equation for Energy Performance Index (EPI), based on the following algebra:

This equation is of the form of the Power Performance Index (PPI) presented later, however in
this case it is in terms of energy and is EPI.
Acceptable models (e.g. SAM) inherently include “compensation factors” as part of the model.
It is necessary to input actual weather data in a climate file. In the case of SAM, actual hourly
data for GHI, DNI, DHI, dry-bulb temperature, and wind speed can be incorporated into TMY3
format file and read by SAM. Other parameters included in the TMY3 file, such as dew-point,
relative humidity, pressure, and albedo can be assumed to be acceptable from the original TMY3
file for the specific location.
The rate of change of the compensation factors affects the time frame over which the summation
is performed.

Instructions for Calculating Long-Term Energy Performance Index

EPI = Actual energy output / Expected energy output

When calculating the expected energy output, System Advisor Model (SAM), or equal, requires
actual weather conditions to be formatted in a Typical Meteorological Year format.

14
Procedure:
1. Download a TMY3 file in the vicinity of the PV array.
2. Click and open the function “Create a TMY3 File”
3. Obtain one year of hourly data for actual weather conditions at the PV array.
4. In order to calculate the Direct Normal Irradiance (DNI), the Direct Insolation Solar Code
(DISC) model developed by Dr. E. Maxwell of the National Renewable Energy
Laboratory, available on-line can be used.
5. The Diffuse Horizontal Irradiance (DHI) was also calculated through DISC data. Using
the DNI, and ϴz zenith angle, calculated by DISC, and the relationship between GHI, the
direct horizontal irradiance (dHI), and DHI, were able to be calculated.

Where:

6. Using the SAM Create a TMY3 Function, create a TMY3 file for your PV system.
7. Input system design characteristics and assumed derate factors into SAM and calculate an
expected hourly generation (kWhAC).
8. Calculate an hourly performance index using the measured energy generated and
expected energy generation from SAM.
9. Apply a filter removing all hours where less than a threshold was generated.
A plot of EPI is provided below from the 600kW PV system applying the above method. It
shows a potential performance problem in late summer that could be investigated, such as
soiling.

15
Figure 2.4: Daily EPI-SAM for One Year Using Actual Weather and SAM

Energy Performance Index (EPI) - Regression

Method:
The Energy Performance Index (EPI) is calculated using a polynomial regression analysis
method to develop an equation relating actual irradiance, temperature, and other relevant
parameters (such as inverter efficiency) to the actual AC energy output at each sample time. The
general equation shown below has four unknown coefficients, and in principle they can be
determined with four equations. Considering that each row of data represents each hour of
operation with values for each of the input and output parameters, therefore there is enough data
to use statistical methods to find the “best fit” equation for a combination of the input parameters
in the regression equation. Deviations between the calculated expected AC output and the actual
output are called residuals and are minimized as the model is improved.
General regression equation:
AC Output Energy = A + Temp×Irrad×B + Irrad×C + Irrad2×D
Coefficients (A, B, C, D) are determined by pseudo-inverse matrix operations in Excel or
MatLab. Automated processing of the regression method is available in statistical programs such
as MiniTab, JMP, SPSS, etc.
Data is needed from an “equation training period” where it is assumed that the system operates
properly and data collected for use in developing equation coefficients. The data needed
consisted of the actual metered hourly kWhAC output, and actual hourly weather (GHI or POA
irradiance, ambient temp, wind, inverter efficiency, etc) input.
An advantage of using the regression analysis method is that an accurate PV model (e.g. SAM)
and correct derate factor are not needed.
The value for EPI is calculated based on actual kWhAC / Expected kWhAC from regression
model using coefficients from actual hourly weather and hourly energy output data over the
previous year.

16
The process to calculate EPI is:
1. Obtain hourly metered kWhAC for assessment period.
2. Sum hourly kWhAC, for each day using Excel SUMIF.
3. Calculate DNI, DHI, DiffHI from GHI using NREL DISC, or use POA data if available.
4. Calculate POA irradiance using Isotropic Sky model.
5. Calculate coefficient matrix using polynomial equation.
6. Use matrix pseudo-inverse to calculate hourly kWhAC .
7. Sum hourly kWhAC for day using Excel SUMIF.
8. Calculate EPI for day, plot daily trend.
Using the regression analysis method, the estimated daily energy can be calculated for
comparison to the actual, Figure 2.5 shows how the calculated and actual compare.
Figure 2.5: Plot Showing Agreement of Regression Equation

Different variables in the general regression equation were tried, such as inverter efficiency
which played a role to reduce the uncertainty. By analyzing weather data, it was found that at
high ambient temperatures the inverter shuts down even though the insolation was optimal for
high power output. Hourly data was used to determine the regression coefficients from the
general regression equation. The POA irradiance was calculated using the NREL- DISC program
using GHI data.

17
Using the general regression model and adding the inverter efficiency to the equation as a new
parameter reduced the uncertainty. With this new parameter, the uncertainty with a GHI greater
than 800 (W/m²) is 4.1%.
Figure 2.6: Plot Showing Result of Regression Method with 4% error bars.

Quarterly data was also used to see if any anomalies or trends existed. Quarterly data would be
used to define a regression equation for a particular season. Further work is needed on this topic
to fully assess its usefulness.
The same technique and method was used for 15-minute data. The reduction of averaging over a
longer period of time (for an hour) was the motive for using 15 minute data so that there would
be less averaging involved. For 15-minute data there were more wild points to be considered,
however, by taking into account more variables, 5% uncertainty was achieved.
The data has a range of points. There are more than 10,000 data points which have been graphed
on this chart and there are many more wild points that need to be taken into account. Yet, the
regression was able to predict the outcome within 5% of the actual power output.
For the year of 2011, the chart below shows the EPI obtained. The confidence is high in this case
since the generic model described above was altered to include more variables.
P = A + T H B + H C +H²D + TE +N F +T NG +DH

Where T = Temp, H = Irradiance, N = Inverter Efficiency, D = Humidity.

18
Figure 2.7: Plot Showing Wild Points Adversely Affecting EPI .

Data Quality Issues and Uncertainty

Methods to reduce wild-points and condition data for analysis were applied. One area that can
be improved is the manner in which data is collected or monitored. Currently the provided data
is an average of data during a one hour time period. In doing so the hourly averaging of data
underestimates the actual energy production during high irradiance conditions. This occurs due
to averaging the fluctuations of irradiance over an hour. With large fluctuations the power
generated will adjust quickly, however the module operating temperature will adjust slowly and
remain at a lower temperature.

Anomalies were found as illustrated in Figure 2.8, whereby irradiance, cell temperature, and
system output varied counter intuitive to known PV principles.

19
Fig. 2.8 Data Anomalies Which Are Inconsistent With PV Operating Principles:
Between 12:20 and 12:30 - Irradiance decreases, cell temp constant, power increases
Between 13:10 and 13:30 - Irradiance decreases, cell temp increases, power increases
Between 13:50 and 14:00 - Irradiance large decrease, cell temp small decrease, power increases

Uncertainty estimates for the measured data were based on literature consensus. Uncertainty for
calculations was based on principles of propagation of uncertainty, such as using square-root-
sum-of-squares combination when products were calculated.
Power Performance Index (PPI)
The Power Performance Index (PPI) is the instantaneous actual AC kW power output divided by
the instantaneous expected AC kW power output. The instantaneous expected AC power
depends on many factors, including the instantaneous irradiance and cell junction temperature,
the module technology including STC ratings and spectral and angular response, and the derate
factors. The actual irradiance absorbed by the module cells (referred to as “effective irradiance”
by Sandia) can depend on a number of factors, including the POA irradiance just above the glass
surface, incident angle, glass coatings, soiling, encapsulant, etc.

A desirable module temperature measurement results in the average cell junction temperature
20
across the array under test. The average cell temperature depends on a number of factors,
including ambient temperature, irradiance, wind speed and direction, mounting geometries, etc.
Uncertainty results from the specific model used to calculate the expected power. Neglecting
some of the factors mentioned above increases uncertainty, but generally simplifies the
calculations and measurements. A detailed PPI analysis could be performed using SAM or other
PV design software to calculate the expected power output considering all relevant factors. A
simple model for the calculation of expected output uses the rated DC STC power (P) times
adjustment factors (Ks) which include instantaneous irradiance and temperature, and is called the
PKs method in this article. The actual power is then compared to the resulting expected power in
the PPI ratio.

Latency between the irradiance and temperature measurements and the actual power reading
should be minimized. Instantaneous measurements are ideal. If irradiance and temperature
measurements are taken manually, it is important to carefully timestamp actual power readings
and irradiance and temperature readings and note how steady the values are so that the actual
power value is correlated to the actual irradiance and actual temperature values. Experience has
shown that apparently clear sky conditions can result in significant variations of irradiance over a
short time. It should also be noted that the uncertainty in the actual power reading shown on an
inverter can vary from inverter to inverter. A revenue grade AC power meter is usually the best
method.

Instructions for Calculating Power Performance Index (PPI) using the PKs method
1. Visually inspect system - Determine as-built configuration, identify conditions affecting
performance, estimate typical derate factors per PVWATTS description or similar
documentation and combine to obtain derate K factor (KDerate).
2. Measure Plane of Array (POA) irradiance. If only horizontal data is available (GHI),
convert to POA using NREL DISC spreadsheet to calculate DNI, DHI and use Isotropic
model to convert to POA irradiance. Note that converting from GHI to POA will
introduce error into the irradiance measurement, especially at steep incident angles seen
early or late in the day. The Isotropic model formula is:

3. Calculate irradiance K factor, KIrrad, from:

4. Measure module backside temperature and add an offset to account for temperature
difference between backside and cells, such as 3°C × KIrrad, per King 2004 paper shown
below. If backside temperature is not available, you can measure ambient temperature
and calculate cell temperature using Sandia model or NOCT value on module datasheet

21
 using one of the following formulas. Note, this method will generally be less accurate
than directly measuring the backside temperature.

or, per Sandia model:

where, ΔT is temperature rise over ambient, such as 3°C.

5. Calculate temperature K factor (KTemp) for temperature relative to STC using the
following formula, where μ is the power temperature coefficient and is a negative
number, such as typically - 0.005/°C.

6. Calculate expected AC output power (kW):

7. Measure actual AC output power (kW) or use inverter displayed value at a time which is
correlated with the irradiance and module temperature measurements.
8. Calculate ratio of measured actual AC power to expected power, define values as Power
Performance Index (PPI)

9. Estimate uncertainty values for measured and calculated values (apply propagation of
uncertainty method using square root sum squares of each relative uncertainty in %).
10. Evaluate PI. If PI = 1.0 ± uncertainty, short-term system performance is acceptable,
proceed to Long Term Assessment.

Uncertainty of PPI
As noted above, the uncertainty associated with the PPI is highly dependent on the model used to
determine the expected power and the methods used for determining instantaneous irradiance
and module cell temperature. The PKs method trades-off uncertainty for simplicity. This
method results in an uncertainty of 10-15%. More sophisticated measurements and models will
reduce the uncertainty of the expected power and therefore reduce the uncertainty of the PPI.
Some industry tools and models have been shown to have uncertainty less than 5%. To reach
22
such accuracy levels requires care. For example, a better method for determining the average
cell junction temperature across the array is to use the Voc of the array as described in IEC 904-
5. This is generally a better method than backside temperature measurements plus offset or
calculations that take into account wind speed because in general it is nearly impossible to
determine the typically non-uniform distribution of wind flow over an array. It should be noted
that the Voc method has limitations at low irradiance values.
Accurate irradiance measurements can be achieved by using a matched reference cell or a model-
corrected reference cell oriented in the POA. This will give a good estimate of the effective
irradiance actually absorbed by the cell, taking into account angular and spectral response, glass
coatings etc and will generally be superior to inexpensive irradiance meters. These kinds of
irradiance meters are fairly accurate when oriented directly at the sun to give the direct normal
irradiance, however direct normal readings are only valid when the sun is directly normal to the
plane of array which in practice is only a few times during the year. Therefore angle-response
effects are not accounted for like when a reference cell is used oriented in the POA.

Conclusion and Recommended Performance

Assessment Methods
Literature review and discussions with industry experts suggested focusing on the following four
metrics:
 Power Performance Index (PPI) of actual instantaneous kW AC power output divided by
expected instantaneous kW AC power output.
 Performance Ratio (PR) of final yield divided by reference yield over an assessment
period.
 Performance Ratio with final yield corrected for cell temperature (CPR) over an
assessment period.
 Energy Performance Index (EPI) of actual kWh AC energy divided by expected kWh AC
energy as determined from an accepted PV model, such as SAM (EPI-SAM), using
actual climate data input to the model over the assessment period, or a regression model
using operating data to “train” the model (EPI-REGRESSION) resulting in reduced
uncertainty since derate factors are not needed.
The three primary objectives for performance assessments of existing systems and the associated
recommended metrics are listed below. A guideline summary is provided in Table 2.2.
 Monitoring of a specific PV system to identify degraded performance and need for
maintenance based on condition. Use EPI metric and trend EPI for the specific system.
 Commissioning, re-commissioning, troubleshooting, or assessment after major
maintenance. Use PPI and EPI metrics.
 Determination of specific industry parameters, such as Yield or Performance Ratio, to
allow comparison of systems in different geographic locations for design validation or

23
 investment decisions. Use PR, CPR and/or EPI depending on the level of effort and level
of uncertainty.
Additional work is recommended to develop specific procedures for each of the four metrics
summarized above and for making Excel spreadsheets available for general use. Additional
long-term data should be analyzed to investigate the ability of metrics to meet the stated
purposes, and to determine best practices for obtaining reliable inputs with currently available
industry products such as monitoring systems and IV curve tracers. An industry standard would
also be useful to improve consistency in calculating and interpreting these performance metrics
across the industry.
Acknowledgements
Contributors to the work summarized in this report included SJSU students: Harikrishna
Patadiya, and Uriel Rosas, and SunSpec Performance Committee members: Willard MacDonald,
Laks Sampath, Mike Balma, Tim Keating, and David McFeely.
The SunSpec Alliance wishes to acknowledge SolarTech, where this project was initiated by the
same team, and originally published Version 1 of the document. The project was moved to the
SunSpec Alliance in 2013.

24
 Table 2.1- Commonly Used Performance Metrics
METRIC CALCULATION REFERENCE
Yield kWh / kWDC STC NREL/CP-520-37358
Performance Ratio (kWh/ kWDC STC ) / (H/GSTC) IEC61724
Performance Ratio kWh / (sunhours × area × efficiency) SMA
(EActual / EIdeal) * 100%
Performance Ratio SolarPro, Taylor & Williams
EIdeal is temp. and irrad. compensated
Specific Production MWhAC / MWDC STC SolarPro, Taylor & Williams
(100 * Net production / total incident
Performance Ratio NREL/TP-550-38603
solar radiation) / rated PV module eff.
Performance Factor ISC,G*RSC*FFR*ROC*VOC,T Sutterlueti
Performance Index kWmeasured / kWexpected SolarPro, Sun Light & Power
Actual Power / (Rated power * irrad adj.
Performance Index * temp adj * degradation adj * soiling adj Townsend
* BOS adj)
Output Power Ratio kWmeasured / kWpredicted SolarPro, Sun Light & Power
Output power kW > CF-6R-PV Table CEC Commissioning
Output power kW > 95% expected SRP Arizona Utility
Specific Production MWhAC / MWDC-STC SolarPro, Taylor & Williams
Acceptance Ratio kWactual / kWexpected Literature
Inverter comparison kWh of multiple similar inverters Qualitative
String comparison Imp, Vmp of multiple parallel strings Qualitative
Utility billing Monthly comparison Qualitative
Performance Ratio,
(kWh/ kWDC *KTemp) / (H/ GSTC) Proposed in this report
temp. comp. (CPR)
Energy Performance kWh AC actual / SAM AC Expected
Proposed in this report
Index (EPI) using actual weather data
Power Performance
kWAC / (kWDC *KIrrad*KTemp*KDerate ) Proposed in this report
Index (PPI)

25
Sample of Excel Spreadsheets to calculate Performance Ratio (CPR)
PR ADJUSTMENT FACTOR DEVELOPMENT
Arizona Game & Fish 191 kW system live site data used to find PR adjustment factors to result in a PR value that has
minimal variation with time, irradiance, and temp, when in normal condition. Decrease in PR would then be due to degradation.
Plots on next tab shows results. System DC Rated Power (kW) = 191 Used only for comparison:
Input Site Data = Cut-off irrad (kW/M^2) = 0.75 NOCT (C)= 47
Output for plots = Power Temp. Coefficient (W/C) =-0.005 Wind (m/s) = 2

Daily PR
Average
using Daily PR
Hourly Measured Total Total Sum of
Hourly AC Average Total Power using
Irradianc Average Daily AC Daily Power
ID y m d t Energy Ambient Time Week Hourly PR Daily times Power
e Cell Energy Insolation times
(kWh) Temp. (C) Energy Hourly PR Weighted
(kW/M^2 Temp. (C) (kWh) (kWh/M^2) Hourly PR
and Average
)
Insolation
1 2011 2 13 0:00 0 0.004 12.664 7.747 0.00 1 0 0 0 0 0 0 0
2 2011 2 13 1:00 0 0.004 12.711 7.649 0.04 1 0 0 0 0 0 0 0
3 2011 2 13 2:00 0 0.004 11.68 6.487 0.08 1 0 0 0 0 0 0 0
4 2011 2 13 3:00 0 0.004 11.069 6.128 0.13 1 0 0 0 0 0 0 0
5 2011 2 13 4:00 0 0.004 11.33 6.604 0.17 1 0 0 0 0 0 0 0
6 2011 2 13 5:00 0 0.004 11.268 5.119 0.21 1 0 0 0 0 0 0 0
7 2011 2 13 6:00 0 0.004 10.469 3.426 0.25 1 0 0 0 0 0 0 0
8 2011 2 13 7:00 4.24 0.152 9.838 6.476 0.29 1 0 0 0.14595 0 0.618827 0 0
9 2011 2 13 8:00 41.732 0.549 11.617 18.17 0.33 1 0 0 0.39791 0 16.60556 0 0
10 2011 2 13 9:00 85.684 0.851 15.261 30.12 0.38 1 0 0 0.527091 0 45.16328 0 0
11 2011 2 13 10:00 125.876 1.069 18.872 46.435 0.42 1 0 0 0.616441 0 77.59508 0 0
12 2011 2 13 11:00 133.224 1.155 21.793 52.784 0.46 1 0 0 0.603851 0 80.4474 0 0
13 2011 2 13 12:00 129.68 1.117 23.064 52.186 0.50 1 844.996 7.584 0.607782 0.583265 78.81712 505.3351 0.598033
14 2011 2 13 13:00 111.624 0.931 23.454 43.163 0.54 1 0 0 0.627665 0 70.06247 0 0
15 2011 2 13 14:00 94.496 0.76 22.913 35.236 0.58 1 0 0 0.650893 0 61.50675 0 0
16 2011 2 13 15:00 79.572 0.641 23.137 34.701 0.63 1 0 0 0.649832 0 51.70844 0 0
17 2011 2 13 16:00 33.172 0.284 22.582 25.096 0.67 1 0 0 0.611318 0 20.27863 0 0
18 2011 2 13 17:00 5.696 0.067 20.232 16.015 0.71 1 0 0 0.444441 0 2.531536 0 0
19 2011 2 13 18:00 0 0.004 18.415 11.741 0.75 1 0 0 0 0 0 0 0
20 2011 2 13 19:00 0 0.002 17.76 10.554 0.79 1 0 0 0 0 0 0 0

26
Sample of Excel Spreadsheet to calculate Energy Performance Index (EPI-SAM)

CALCULATED RESULTS FROM SAM MODEL OF 600KW SYSTEM

SAM results for modelled 600kW system with actual weather data
SUMIF FUNCTION:
Expected
Incident Incident
AC Power System Cell Hourly Actual
SAM AC Total POA Incident Radiation Daily Daily
(kWh), Output Temperature System
Hour Day Output (kW/m2) Total POA (kWh) Day Actual Expected EPI
Monthly (kWh) ( C ) From Output
nonzero From (W/m2) From kWh AC kWh AC
from SAM From SAM (kWh)
SAM SAM
SAM
36061.9 1 1 0 0 0 0 0 0 0 1 315.9329 280.3982 1.126729
46792 2 1 -0.13108 0 0 0 0 6 0 2 430.5748 346.1431 1.243921
56134 3 1 -0.13108 0 0 0 0 5.5 0 3 1014.078 948.5689 1.069061
89474.7 4 1 -0.13108 0 0 0 0 4 0 4 1392.243 1379.152 1.009492
101346 5 1 -0.13108 0 0 0 0 3 0 5 1343.986 1404.672 0.956797
106309 6 1 -0.13108 0 0 0 0 2.5 0 6 1280.172 1254.247 1.02067
117638 7 1 -0.13108 0 0 0 0 2.5 0.08 7 615.9333 594.02 1.03689
105188 8 1 2.1934 2.1934 0.020889 20.8889 72.0133 3.54048 7.716667 8 551.3967 502.4821 1.097346
82431.3 9 1 29.3149 29.3149 0.074904 74.9042 258.228 6.43809 36.35 9 1025.821 964.2484 1.063855
53676.3 10 1 51.1864 51.1864 0.124594 124.594 429.531 12.9768 57.5 10 1308.351 1387.622 0.942873
41686.7 11 1 65.7219 65.7219 0.158559 158.559 546.623 14.1003 70.64286 11 377.4438 336.7424 1.120868
23056.1 12 1 33.1362 33.1362 0.085702 85.7017 295.452 13.7125 37.08333 12 1381.56 1366.975 1.010669
13 1 45.4009 45.4009 0.114619 114.619 395.142 16.5207 51.55 13 501.4148 424.0853 1.182344
14 1 29.192 29.192 0.077618 77.6176 267.583 16.9421 29.66 14 1069.255 1076.823 0.992972
15 1 22.3394 22.3394 0.061528 61.5277 212.114 17.4332 23.85 15 1386.507 1356.834 1.021869

Daily EPI-SAM
1.6
1.4
1.2
1
0.8
Daily EPI
0.6
0.4
0.2
0
0 100 200 300 400

27
Sample of Excel Spreadsheet to calculate Energy Performance Index (EPI-Regression)

28