Volume 1 photovoltaic solar energy 1 38 – performance monitoring

Volume 1 photovoltaic solar energy 1 38 – performance monitoring Volume 1 photovoltaic solar energy 1 38 – performance monitoring Volume 1 photovoltaic solar energy 1 38 – performance monitoring Volume 1 photovoltaic solar energy 1 38 – performance monitoring Volume 1 photovoltaic solar energy 1 38 – performance monitoring Volume 1 photovoltaic solar energy 1 38 – performance monitoring Volume 1 photovoltaic solar energy 1 38 – performance monitoring Volume 1 photovoltaic solar energy 1 38 – performance monitoring Volume 1 photovoltaic solar energy 1 38 – performance monitoring

N Pearsall, Northumbria University, Newcastle, UK

R Gottschalg, Loughborough University, Leicestershire, UK

© 2012 Elsevier Ltd

Array yield The energy provided by the PV array over a reference yield for a PV system

defined period, expressed per kilowatt of array capacity Reference yield The energy theoretically available from a Energy conversion efficiency The ratio of the energy PV system operating under reference conditions, expressed provided by the PV system to the energy received as per kilowatt of array capacity

sunlight, usually expressed as a percentage Stand-alone system A system that is designed to be the Final yield The energy delivered to the load by the PV sole power supply for a given load, independent of the system over a defined period, expressed per kilowatt of electricity distribution grid

Grid-connected system A system that is connected to the measurement condition at which the efficiency of the electricity distribution grid such that it can export device is determined This is defined as an irradiance of

Peak power The power of a module or array measured at spectrum, and perpendicular incidence of the

Standard Test Conditions, typically denoted as Wp or kWp irradiance

1.38.1 Introduction

The measurement of performance of any energy generation system allows quantification of its contribution to our energy needs,

to determine whether it is operating in accordance with the design expectations and to make comparisons with other energy systems, whether of the same or different types Establishing the performance level can address any operational problems that reduce energy output and improve system design and performance in the future A photovoltaic (PV) system is inherently low maintenance and requires little user input for its operation in comparison with most other electricity generation technologies Thus, it is particularly important to understand how to determine system performance without requiring a disproportionate effort from the user There is also a wide range of system types, each one having its own performance characteristics and typical values

Of course, simply determining the current performance cannot, in itself, tell us if the system is performing in line with expectation We also need to consider how to predict the performance of the system in the location and for the application considered Although much of that prediction is based on how the PV module performs under different conditions of solar irradiation and temperature, we also need to consider other system components, maintenance activities, and performance changes with time in order to measure and analyze system performance

In this chapter, we will discuss the following:

• Methods of expressing performance at module and system level

• Prediction of performance as a function of location and operating conditions

• Measurement procedures for establishing performance and how these are implemented in practice

Comprehensive Renewable Energy, Volume 1 doi:10.1016/B978-0-08-087872-0.00144-X 775

• Performance monitoring for different module and system types

• Challenges in long-term monitoring

1.38.2 Defining Photovoltaic Performance

Before considering how performance is measured in practice, the term must be defined in the context of a PV system and what parameters can be used to express it

PV performance is generally carried out for three distinct categories: modules, grid-connected systems, and stand-alone systems The detailed design of the system is considered in other chapters in this volume, and thus this chapter will only deal with the evaluation of performance Modules are evaluated largely to determine their energy yield and stability, with the output informing system design The outcome is the specific behavior in different operating conditions Grid-connected systems are operated in parallel with the conventional grid distribution system, and the overall energy output is the main parameter of interest Stand-alone systems are designed to meet a specific load and, in general, to be the sole power supply for that load Therefore, it is not the overall energy output that is of interest, but how well the system meets the load or provides the service that is required, for example, the provision of light or pumping of a particular volume of water Therefore, different interpretations of performance are required, including what is measured, how the measurements are analyzed, and how the analyses are interpreted Note that this chapter will only deal with the measurement and understanding of PV system performance for terrestrial systems and will also mainly consider flat-plate systems rather than those used under concentration, although many of the same principles apply

The main parameters used to describe PV performance are efficiency (the ratio of the electrical output to the solar input, expressed

in terms of power or energy), yield (the energy output over a defined period normalized to the module or system rating), performance ratio (PR) (which expresses the energy output of a system compared with the ideal system), and cost per generated unit of electricity The last of these will not be considered in this chapter Efficiency can be determined for individual components or for the whole system and yield can be defined at the module or system level Over the past 20 years or so, the best way of expressing performance has been the subject of considerable discussion, since the output of any system is dependent on the specific operating conditions In this chapter, generally accepted definitions of performance will be used, but with some comments on their limitations and usage It should also be noted that different stakeholders will adopt different definitions as appropriate to their purposes

1.38.2.1 System Efficiency

The efficiency of a PV component or system can be defined in terms of power or energy, where the latter is for a defined period It is usual to define the power conversion efficiency of a PV module under Standard Test Conditions (STC, defined as a solar irradiance

of 1000 W m−2 at normal incidence, an operating temperature of 25 °C, and a defined AM1.5G spectrum [1]) However, as operating conditions change, so does the module efficiency, due to light intensity and temperature effects The energy conversion efficiency, which averages the power conversion efficiency over the range of operating conditions experienced, can be more informative in terms of assessing the performance of a real system

The energy conversion efficiency of any component or system is calculated by comparing the energy output of that component/ system over a defined period with the energy input over the same period In the case of a PV module or system, the energy input is given by the solar energy falling on the PV module or PV array surface, respectively This is usually calculated from the solar irradiation level, expressed in kWh m−2, multiplied by the module or array area, as required, in m2

Energy output

Area  Irradiation where efficiency is expressed as percentage and the energy output is measured in kilowatt-hours The irradiation is calculated for the same period as the energy output is determined (e.g., average daily irradiation multiplied by the number of days in the period) Thus, in order to determine the energy conversion efficiency for a real system, measurements of the energy output of the system and the irradiation falling on the array surface are required

At the module and system level, the efficiency values derived will depend on the efficiency of the individual system components, particularly the PV module and the power conversion equipment (inverter) in the case of grid-connected systems As for the PV module, the energy conversion efficiency over any given period is dependent on the range of operating conditions experienced For a stand-alone system, it is a little more complicated, since the measured efficiency will depend on the detail of the load and the storage component This is because the system does not always work at the maximum power point of the array, particularly when the battery is close to full charge (see Section 1.38.2.4 for further explanation)

1.38.2.2 System Yield

Yield figures describe the energy output over a defined period normalized to the power rating of the PV module or system It is usually expressed in kWh kWp −1 This normalization allows yield values to be compared between modules or systems They are often expressed as either average daily or annual values

The following four yield values are generally considered:

• Module yield The energy output of a single module (DC output)

• Array yield The energy output from the PV array (DC output)

• Final yield The final energy output of the system, the energy delivered to the load (or to the grid in the case of export from a grid-connected system)

• Reference yield The energy output of the system if it operated at rated power levels at reference irradiance conditions for the number of hours that are equivalent to the irradiation received over the period

In order to monitor the yield values, we require a measurement of the energy output and knowledge of the array rating For the array yield, the DC energy output of the array should be measured, although this measurement is not always included in simple monitoring systems The reference yield is more easily calculated by measuring the irradiation falling on the array It can be shown that the reference yield is numerically equal to the irradiation measured in kWh m−2

The overall losses in the PV system can be estimated by comparing the final yield with the reference yield and this is the basis of the PR, which is discussed in the Section 1.38.2.3 Comparison of the array yield and the final yield gives information on the losses

in the balance of systems components (inverter, wiring etc.) Comparison of the array yield and the reference yield determines the capture losses of the PV array, that is, the losses associated with the module operation itself These losses include reduction of the absorption of light due to angle of incidence effects, reflection from the module surface, and dirt accumulation, together with an operating temperature higher than that for the standard measurement conditions

Yield values are location-dependent, although typical values can be determined for a region A final yield value based on the simulation of system operation under average climatic conditions is often used to define the predicted output of a PV system 1.38.2.3 Performance Ratio

The PR of the system is calculated from the final and reference yields as follows:

Final yield

This expression can also be rewritten as

Final energy output

where the final energy output is expressed in kWh, the array rating in kWp, and the total irradiation in kWh m−2 For measured data, the monitoring fraction, expressed as a number between 0 and 1, is introduced in the denominator to allow for missing data The monitoring fraction expresses the proportion of actually measured values to planned measured values in the period PR can be expressed as a fraction or as a percentage and is dimensionless

In order to determine the PR value for a system, only the measurement of the final output and irradiation values in the plane of the array are required, together with the knowledge of the array rating This makes it a fairly straightforward parameter to measure Essentially, the PR value gives a direct measure of the losses in the system by comparing the actual output of the system with that from an ideal system at the same location and under the same climatic conditions Because it takes into account both the system rating and the irradiation conditions, it becomes possible to make direct comparisons of systems with different designs and in different locations and to provide guidance on acceptable PR values A high-quality grid-connected system can be expected to have an annual PR value of 0.8–0.85 If lower values are obtained, the reason for the reduction should be understood and addressed if it is possible to do so Recently, there have been some discussions of whether PR values give an appropriate measure of performance, mainly since the calculations include the power rating of the array (see, e.g., King [2]) The array’s power rating is determined by summing the STC ratings of all the modules connected into array and it is common to use the nameplate rating of the module for this purpose Of course, there is a variation in the power rating arising from the manufacturing process and so there will be a level of uncertainty on that rating value This can lead to a higher or lower PR value than would be expected as a result of the system losses, due to a higher or lower overall module rating than the nameplate value The situation is complicated further if the module power is known to vary with time, for example, the seasonal variation in power output due to light exposure and annealing of amorphous silicon modules In this case, it

is usual for the nameplate rating to be the stabilized value of STC power, and so the system will appear to have a high PR value in early operation before the module output has stabilized Nevertheless, as long as these limitations are understood, it is the authors’ opinion that the PR value remains a valuable parameter for assessing and comparing the performance of grid-connected systems In particular,

it is very useful as an ongoing monitoring parameter, where the change in PR indicates changes in the loss levels in the system Also, as yet, alternative assessment parameters appear to have other disadvantages in terms of direct identification of losses

1.38.2.4 Performance of Stand-Alone Systems

The main judgment of performance for a stand-alone PV system is whether the load receives adequate power levels to allow it to operate for the required amount of time, consistent with the system design specifications Assuming a stand-alone design with

battery storage, the power to the load is assured if the PV array size is large enough to ensure that the battery is at a high state­ of-charge (SOC) at most times and that the battery capacity is large enough to provide power to the load without charging for the number of days required by the climatic conditions In general, the higher the reliability required for the system, the larger will be the array and the battery capacity in the design

At times when the battery is close to or fully charged, the operating point of the array will be moved away from the maximum power point in order to reduce the charging current to an acceptable level and to match, but not exceed, the load This is in contrast

to a grid-connected system, where the usual mode of operation is to try and work at maximum power point at all times Therefore, since the stand-alone system is working away from reference conditions, the PR will be reduced, even though the system is working exactly in line with design expectations and there are no unacceptable system losses

In general, the lower the loss of load probability specified in the design, the more often the system will operate under conditions where the battery is at high SOC and therefore the lower will be the apparent PR This means that systems designed for critical applications, where it is required to have a very low probability that the load will be lost, will have apparently low performance if the

PR value is considered Indeed, a high PR for a stand-alone system only indicates that the user is consuming nearly all the energy that can be produced under normal conditions, running the risk of loss of load under poor weather conditions The situation is not improved if we use system efficiency or yield, since those parameters are also based on energy output rather than service provided Some measure of the availability of power to the load is required, although it is also possible to consider battery SOC as an indicative parameter Several performance parameters have been proposed for stand-alone systems to be used alongside or instead

of PR Some of these take account of the length of time for which the battery is fully charged and/or propose consideration of charge parameters for the assessment in place of energy parameters One of the issues to be addressed is that stand-alone system designs vary substantially, so all parameters tend to have a fairly large acceptable range which makes it more difficult to identify faults Vallvé (Trama Tecno Ambiental, private communication) has suggested the use of the battery index, which represents the number of days on which full charge is reached Values over 30% are considered good, between 10% and 30% marginal and perhaps warranting investigation, and below 10% unsatisfactory A low battery index suggests that loss of load is likely under poor weather conditions and that either the system design is unsuitable or that performance is not in line with expectation The total time

of disconnection is measured and it is also necessary to determine if that disconnection is due to overconsumption (rather than lower than expected generation) in order to determine whether the problem is one of design or usage patterns The charge into and out of the battery is also derived from measurements of the current on both sides of the battery As a bonus, these measurements also give some information in relation to battery aging and the requirements for maintenance

1.38.3 Module Energy Prediction

Today, the prediction of the energy yield of PV modules is carried out in the initial stages of a system design in order to choose between different competing devices on offer The basic building blocks of a yield prediction are illustrated in Figure 1 The ingredients are always a module description, the description of the environment and the mounting data The latter is required as it influences the environment being seen by the device in terms of irradiance and device temperature

There are two main uses of an energy yield prediction, which are distinguished by the environmental data sets being used One is

to predict for a specific site, which is then often called an energy certificate or a yield certificate The second is to use a more generalized prediction that could, for example, be put into a module datasheet, where one would speak of an energy rating as currently being proposed by the International Electrotechnical Commission (IEC) [3] A rating does, however, require a standard module description, a standard set of environmental data, and a standard methodology to calculate the output

Module Data

Environmental Data

Mounting Data

Calculation Method

Energy Yield

Figure 1 General modeling overview

Figure 2 Evaluation of modeling accuracy The figure shows an intercomparison of eight different modeling approaches, and the average deviation and the deviations in the modeling are shown RR1 uses irradiance and temperature effects only; sm, spectral modeling; rm, reflection modeling; tm, modeling

of module temperature; hor, translation of irradiance from the horizontal to the inclined and all the overall methods’ accuracy Adapted from Friesen G et al (2009) Intercomparison of Different Energy Prediction Methods Within the European Project ‘Performance’ – Results of the 2nd Round Robin In Proceedings of the 24th European Photovoltaic Solar Energy Conference [13]

The calculation method determines the structure of the input data There are simplified empirical methods [4–6], methods simulating the current–voltage (I–V) characteristics [7–9], and statistical methods [10] There are some differences between the required inputs, as some models model only irradiance and temperature effects, while others include effects of variable spectrum and angle of incidence Recent intercomparisons have shown that even simple methods only modeling irradiance and temperature behavior can predict the yield to  3% in the case of crystalline silicon devices This is shown in Figure 2, where the results of a recent evaluation of modeling approaches are shown Overall, eight models were assessed The starting point was modeling of irradiance and temperature effects only (indicated as RR1) The bars demonstrate the average agreement, while the error bars indicate the variations of the different modeling approaches If the irradiance and the module temperature are known, then accuracies are expected to be in the range of 3% However, normally one would start with the horizontal irradiance as given by relevant meteorological data sets and calculate the module temperature The categories of hor and tm in Figure 2 represent the addition of these factors One can observe that most thermal models used in this group of participants tend to overestimate the temperature, resulting in an underestimation of power This is due to the fact that hourly data sets, as used for this intercomparison, do not have any information on the variability of the irradiance during the time step Paradoxically, the highest irradiances tend to occur on days with light clouds [11], which results in high irradiance conditions often being associated with variable operating conditions, which then results in the operating temperature being nonlinear with irradiance at higher irradiances [12]; in fact, they appear to be reduced compared with irradiances around, say, 900 W m−2 The calculation of the in-plane irradiance increased the uncertainty in the models significantly It was also noted that when including spectral effects (sm) and reflection effects (rm) into the models, accuracy did not increase The reasons for this are manifold, but, for example, in the case of the spectral modeling the apparent overestimation of energy was balanced with the reflection losses, so overall the model performed in the same way as without these effects The issue is largely related to the available input data The differences introduced by the different calculation blocks (Figure 1) are thus not the drivers in the accuracy of a yield prediction It was also noted during this intercomparison that models tended to perform well for one type of module, but less well for another Thus, the differences observed in professional tools cannot be due to the different calculation and mounting structures There are differences of some percentage points but not the values of >5% deviation that were reported for some commercial packages The module description is a major issue in any performance prediction as the required data are often not available This is due to a number of issues, including that data provided by manufacturers are not consistent, there are uncertainties in the measurement of the different properties and finally the way modeling software parameterizes input data often results in very different device behavior The data available for modeling purposes are often taken from the datasheet At the moment, this can be rather inconsistent The temperature coefficient is measured for one module, the low light behavior for another (potentially by another laboratory altogether), and the STC power is a nominal value As a result of manufacturing variations, modules are classified into different power categories, commonly known as power bins, depending on their measured output This means that the power of a specific module will usually vary slightly from the nameplate power The temperature coefficient is typically also larger for lower power bins than for higher ones, and if the temperature coefficient is not measured for a module of the specific bin, this can result in some deviations which will be visible in the prediction accuracy The work of Friesen et al [13] has shown that the difference of the specific module values for those given on the datasheet is by far the dominating factor in the modeling accuracy One approach to get consistent data is the application of IEC 61853-1 [3] as this requires all measurements to be taken on one module However, this is not a datasheet requirement [14] and thus is not available for all module types

There are also significant deviations between laboratories in the way key parameters are measured This depends on a great many parameters, including the equipment used and the traceability chain in the measurements A European intercomparison [15]

reported that after having carried out an optimization, an agreement between the European test and measurement laboratories of

−6%

−4%

−2%

0%

2%

4%

6%

c-Si 1 c-Si 2 CdTe

Figure 3 Agreement of power measurements in European test laboratories at varying irradiances Data detailed in Herrmann W (2010) Results of the second round robin test of PV-performance Internal Deliverable D1.1 4: 25 [16]

H1 H2 H3 S1 S2 S3 S4 S5 T1 T2 T3 T4 O1 H1 H2 H3 S1 S2 S3 S4 S5 T1 T2 T3 T4 O1

− 0.15%

Manufacturer Program V1 Program X2 Program Y2 Program Z1

Hi eff cSi Std eff cSi Thin Film Oth 20%

15%

−

Hi eff cSi Std

Manufacturer Program V1 Program X2 Program Y2 Program Z1

eff cSi Thin Film Oth

− 0.25%

5%

− 0.30%

0%

− 0.35%

− 0.40%

− 5%

− 10%

− 20%

− 0.50%

− 25%

Figure 4 Comparison of manufacturer’s data with values in simulation programs for the temperature coefficients (left) and low light behavior (right) Graphs taken from Ransome S, Sutterlueti J, and Kravets R (2011) How kWh/kWp modelling and measurement comparisons depend on uncertainty and variability In: Proceedings of the 37th IEEE Photovoltaic Specialists Conference Seattle, WA: IEEE [17]

1.5% was achieved between the majority of test houses for crystalline silicon (counting all outliers, this would be closer to 3%) Thin films showed a deviation of about 5% One should keep in mind that this is carried out after one optimization step and without this the agreement is significantly reduced This rating is the uncertainty in the normalization The measurement uncertainty

in the parameters such as low light behavior and temperature coefficient are even more significant The change in the agreement between the different laboratories in one round robin intercomparison is shown in Figure 3 It should be noted that two groups, the one with the maximum measurement and the one with the minimum measurement at STC, did not participate in the low light behavior measurements Thus, it is surprising that the measurement at 100 W m−2 deviates by as much as −4% to +5% [15] Clearly, this will affect the performance of any modeling routine, especially when modeling areas with high contributions of low light behavior The difference in the temperature coefficients was even larger, about 15%, which will have a significant effect on accuracy in high-temperature environments Thus, the measurement uncertainty is a major factor in energy prediction

The input data are then parameterized to make them useful for simulation codes A major reason for the differences in the modeling results is in the way simulation codes translate the datasheet values into the module description for the module Ransome

et al [17] compared the data used by a number of commercial software programs in terms of how well they match the manufacturer’s datasheet The findings are summarized in Figure 4 The temperature coefficient, shown in the left graph, varies

by about 20% for standard silicon devices This is actually within the measurement uncertainties, so could be justified by using different input data sets In the case of thin-film devices, much more significant deviations can be seen The low light behavior is defined here as the ratio of the efficiency at 200 W m−2 to the efficiency at 1000 W m−2 The differences between the modeling programs and the datasheet values can be quite surprising The relative differences are multiples rather than percentage values Again

−6.0%

−4.0%

−2.0%

0.0%

2.0%

4.0%

6.0%

Irradiance [W/m2]

STD Min Max

thin-film devices fare worse in terms of agreement than crystalline devices In areas such as the United Kingdom, where about 30%

of incident irradiance is at levels below 300 W m−2, this will easily cause >5% variation in the energy yield prediction

The other main contribution to the uncertainty is the environmental data It has been shown that even simulations using the same data source can result in rather different outcomes An obvious reason is how horizontal input data are translated into in-plane irradiance The time resolution can also influence the outcome of any simulation step In addition, some devices show a rather significant spectral effect and/or idiosyncratic angle of incidence effects, which often will not be covered by simulation codes One should also keep in mind that the modeling can only be as good as the input data Modeling the spectral irradiance, for example, is either carried out using software codes or spectral irradiance measurements The latter are few and far between, and thus there is a reliance on spectral models In an intermediate step, one can use empirical correlations [18], but that requires longer-term measurements Thus, one typically uses SMARTS [19] or similar codes, but these are not really designed for non-clear days, and thus, modeling the spectrum might actually introduce a bias into any simulation On an annual basis, the angle of incidence and spectral shifts tend to balance In shorter time periods, very significant modeling errors can be seen

Module energy prediction can be summarized as being within 3% accuracy for crystalline silicon devices, if module characteristics, temperature, and in-plane irradiance are known The key uncertainties are in the determination of these characteristics Specific characteristics of unusual devices are more difficult to capture and need further research In particular, if short-term prediction is required, for example, in the case of condition monitoring, the ‘secondary’ effects need to be evaluated in much more detail

1.38.4 PV Systems – Performance Prediction

All the statements of the module energy prediction apply, as this tends to be the backbone of the system energy prediction Most prediction methods will carry out the module energy prediction and then multiply the current and voltage results to account for parallel and series connections The effects of the balance of system (BOS) components are then considered

The measurement of system energy output and other related parameters, such as irradiation received and various voltage and current levels, can determine system performance, but the judgment of performance quality depends on whether this matches with expectation Performance is a function of location, mainly relating to the climatic conditions and the detail of system design, including array orientation, component selection, component matching, and external factors such as shading Therefore, it is important to have a reliable performance prediction against which to compare the measured values

Performance prediction is usually carried out using specialist software tools, either proprietary to the company concerned, available for purchase, or in some cases, available without charge In the latter category, it is possible to obtain a simple prediction of system output for a given location from the Photovoltaic Geographical Information System (PVGIS) web site developed by the Institute of Energy of the European Commission’s (EC) Joint Research Centre [20, 21] It is also possible to obtain performance predictions from a range of inverter manufacturers via their web sites More complex design software, including the possibility to include external shading, select specific components, and modify embedded assumptions, can be purchased, with well-used examples being PVSyst [7] and PV*SOL [22] There is a wide range of PV system simulation software available and the information given here should not be taken as a direct endorsement of the examples provided, although they are both popular and well-used programs Such simulation programs will provide estimates of efficiency, yield, and PR that can then be compared with values achieved in practice

Nevertheless, whatever simulation software is used, it needs to be recognized that the predictions relate to average conditions and use historical weather data Therefore, when comparing with measured values, it is vital to consider how the irradiation level received relates to that used in the prediction process It is also necessary to consider the accuracy with which both the measured and simulated values can be determined

1.38.5 PV Module Performance Monitoring in Practice

The aim of module performance monitoring is, in the majority of cases today, for the differentiation between different modules, that is, demonstration of ‘superior’ energy yield of certain products These lists can be produced ‘independently’ or openly by the manufacturer A second common use is for long-term performance investigation, that is, durability and stability of products Third, monitoring data are often used for the validation of energy prediction methods, or as input into these All these uses have in common that high measurement accuracy is required The section will review common arrangements in terms of measurement systems and the main issues encountered

There are two different setups commonly in use for module monitoring, which are depicted in Figure 5 The left graph in Figure 5

shows a typical system using a multiplexer-based system where I–V traces are taken at fixed time intervals An example of such a system is documented [23] These systems use high-accuracy measurement components and normally four-quadrant power supplies using a Kelvin connection to minimize the influence of any transmission losses The cost of measurements is very high, and thus, normally the measurements are multiplexed and measurements are taken in a serial fashion The right graph in Figure 5 depicts the other extreme, a system which tracks the maximum power point all of the time, as, for example, analyzed by Dittmann et al [24] The measurement components are typically of lower quality than the single measurement system approach, but measurements are taken in a parallel fashion In reality, both system types have started to move towards the middle ground Most measurement maximum power point trackers (MPPTs) now have features to trace the I–V characteristic of the attached device The multiplexer systems mostly have the feature

Figure 5 Typical module monitoring layouts The left graph shows a typical multiplexed system, while the right graph shows one system with multiple maximum power point trackers (MPPTs)

Normalization 3.00%

Measurement uncertainty 0.50%

Module operation 1.00%

Selection 3.20%

Environmental variations 1.20%

Figure 6 Contributions to the comparability of module measurements on a single site

for a load to be present, as it has been shown that otherwise some devices might exhibit unrealistic aging [25] Both approaches are used throughout the world and measurement accuracy depends as much on the circumstances around the measurement system as on the measurement system itself

These monitoring campaigns tend to be mission-critical for manufacturers and thus tend to be frequently overinterpreted, by making rather small differences in energy yield seemingly significant The uncertainty in the kilowatt-hour measurement tends to be rather low [23, 24], but that does not mean that the comparability of the devices is good enough In many respects, the framework of the monitoring is much more important for a good intercomparison Results are also typically compared on a kWh/kWp basis, which also has an influence on the comparability of devices A typical intercomparison would have less than 5% comparability for the specific yield of a module (as illustrated in Figure 6)

The electrical measurements only account for 0.5% in the overall comparability The two major sources of uncertainty are the normalization and the selection The simulations used for Figure 6 are based on a single-diode model They allow all possible parameters

of a PV module within the power bins, which clearly is an oversimplification This would need to be simulated based on the analysis of a specific power production statistic to come to a more precise number A contribution of maybe half of this would be expected (as quality assurance limits the parameter space significantly), but this would still result in a contribution of around 1.5% to the comparability

Switchgear (MUX)

4-quadrant

power supply

SCC block

A/D card Multiplexing

V, Tmod, I Switching control for MUX Measurement PC

PV devices with PT100 at back

Voltage control Voltage sense

Serial cable

Current monitor Current sense

PV devices with PT100 at back

Vcontrol (MPPT) Sense I,V DAQ

Vcontrol (MPPT) Sense I,V DAQ

Vcontrol (MPPT) Sense I,V DAQ

Measurement PC

The most significant uncertainty in outdoor monitoring is actually the calibration of the module power It has been shown in

Figure 3 that the comparability of EU test houses is around 3%, and thus any power used for normalization will have such an uncertainty It is now debatable what peak power value to use Good practice would be to measure all devices in one location (i.e., one test house for all devices) as this reduces the uncertainty in the power calibration to the repeatability of the test house which typically is in the sub-1% range Should one use the measurements supplied in the factory flash test, it should be kept in mind that the manufacturer’s measurement uncertainty is at least 1% higher than that of the test house, as a transfer of the standard has to

be carried out If one uses the nameplate (or datasheet) value, the uncertainty to be used for the intercomparability is the manufacturer’s measurement uncertainty (as the manufacturer is selecting the modules) plus the allowable bin width

Environmental variations in Figure 6 indicate that even in the same location, not all modules might see the same environment Modules in the center of a test field will typically operate at slightly higher temperatures, as they will not experience as much windchill Modules will receive slightly different amounts of light due to small variations in the viewing angles and/or ground reflection These variations are extremely difficult to eliminate, and thus this represents a minimum value for comparability The final item is that different module operations can contribute to the absolute comparability Different cabling lengths can be significant if no Kelvin connection is made MPPTs do not always find the appropriate operating point as they are affected by instability of the environment and also are not 100% effective to find the MPP even in steady-state conditions

Module measurements tend to be carried out on highly instrumented measurement systems, thus the choice of the sensor is not so much of an issue here as it is for PV systems monitoring Typically, one would have a high-quality pyranometer and supplementary reference devices, preferably in a redundant setup If the monitoring data are to be used for energy prediction as well, it is advisable to characterize the incident irradiance further, for example, beam/diffuse irradiance, incident spectrum, and angular distribution of the irradiance A large number of extensions can be added, which all have in common that they are very costly

1.38.6 PV System Performance Monitoring in Practice

It is straightforward to define what measurements can be made in order to assess PV system performance, but less obvious what measurement should be made in any specific case The choice of parameters to measure, sensors and sensor positioning, frequency

of data measurement and storage both the nature and frequency of the data analysis depend on the system type and the underlying purpose of undertaking performance monitoring

The procedures for PV system monitoring were initially developed to determine the performance at installation and over a short period thereafter, typically up to 2 years The main purposes of the monitoring were as follows: (1) the verification of the system quality especially where the installation was funded from public sources (e.g., national government, EC) and (2) the collection of information on the specific operating characteristics in order to allow lessons to be learned for the design of future systems One of the first documents to address the requirements for system monitoring was produced by the EC Joint Research Centre at Ispra, Italy, for use in EC-funded PV demonstration projects [26] and included details on data format to allow collection of all performance data for comparison and system development Subsequently, an international standard on PV system monitoring, IEC 61724, was adopted [27] This document has a similar approach, being mainly concerned with the establishment of performance at the beginning of operation

In the last 15 years, there has been a major shift to a situation where almost all PV installations are now commercial in nature and for the purpose of delivering electrical energy to the user As a result, the purpose of monitoring has also changed from simply providing information on system performance issues to ensuring that the energy provision is as high as possible and in line with the expectations from the design stage The monitoring process reflects the fact that the energy produced from the system now has a financial value This means that the monitoring approach, in terms of what is measured, how rapidly any faults or problems can be identified, what magnitude of loss can be detected and the funds allocated for the monitoring process, depends on the value of the generated energy or the service for which it is used As part of the EC Framework Programme 6 project PERFORMANCE, the European guidelines for PV system monitoring and assessment have been updated in an effort to reflect this change in focus and, especially, the need to consider monitoring and analysis over the full lifetime of the system [28]

In this section, we will first consider what can be measured and some principles of how faults or losses can be determined We will then consider how the monitoring is implemented in practice

1.38.6.1 Measurements and Sensors

The measurement of any quantity requires a sensor that will provide an output that varies as that quantity varies Ideally, the sensor will have a linear response to the variation of the parameter being measured Otherwise, some signal conditioning will be required

to convert the sensor output to the correct value Sensors vary in terms of accuracy, calibration requirements, mode of usage, and cost The accuracy of the measurement has a direct impact on the ability to detect operational problems, since it is not possible to detect changes in performance that are smaller than the uncertainty in the measurement of the performance parameter itself The output of the sensor can be displayed directly, but it is usual to record the outputs for later analysis This also allows the use of historical data to determine performance trends

The frequency with which a given parameter should be measured depends on how that parameter is expected to vary In general, the more variable the parameter, the more often it needs to be measured to ensure that representative values across the required

period are obtained However, the more the measurements made, the greater are the storage capacity required and the analysis effort However, given the advances in data storage in recent years, the issue of storage capacity is no longer a particular constraint and it is often useful to have additional values to consult when trying to understand the loss mechanisms

The most commonly measured parameters for a grid-connected PV system are those given below (with the usual units):

• Global irradiance in the array plane (W m−2)

• Ambient temperature in the shade (°C)

• Module temperature (°C)

• Array output voltage (V)

• Array output current (total) (A)

• Inverter/rectifier AC power (kW) or cumulative energy output (kWh)

• Power to the utility grid (kW) or cumulative energy exported to the grid (kWh)

• Power from the utility grid (kW) or cumulative energy imported from the grid (kWh)

Where the PV system consists of several subsystems, it is useful to make AC measurements for all or a selection of the subsystems Where the PV array consists of a number of subarrays, it is useful to make DC measurements for all or a selection of the subarrays This allows the comparison of the performance of the subsystems or subarrays, which gives a powerful tool for the identification of losses when one subsystem or subarray has a lower output than the others

In terms of defining system performance, the two most important parameters are the global irradiance in the array plane and the cumulative energy output from the inverter, with the other measurements providing information to aid the understanding of the values obtained for overall performance All systems will have some measurement of energy output, although not all have irradiance measurements due to the expense of installing a local sensor Solar measurements can also be obtained from nearby sites (e.g., meteorological stations) or from satellite data Many of the monitoring services make use of satellite data for small systems where it is not economic to install a separate sensor This gives a reduced accuracy in comparison with a local sensor and thus has an impact on the magnitude of problems and the losses that can be identified, but provides a service that could not be accessed otherwise

1.38.6.2 Monitoring in Practice

The way in which performance monitoring is implemented in practice depends on the type and size of the PV system and the purpose of the monitoring itself For most systems, the purpose is to establish the energy output and to determine if there are any problems or issues that need to be addressed The way in which that is done depends on the value of the electricity being generated, the financial arrangements in regard to the installation, and the speed with which any problems need to be identified

For some systems, where they are associated with research and development activities, more complex monitoring is justified since the purpose is to obtain detailed performance data under a range of operating conditions (see also the previous discussion on module performance comparisons) The information may be used for product development, product comparison, or system design improvement Since these are specialist monitoring systems and therefore vary considerably in scope, the remainder of this section will concentrate on the monitoring of commercial systems

For the majority of systems, especially those of small to medium size, the inverter performs the monitoring function and most major manufacturers provide options for measuring, storing, and displaying monitored data This can include the ability to compare subarrays or subsystems and calculate most of the main performance parameters Inverter-based monitoring systems do not usually incorporate solar irradiance sensors as standard, but some provide the opportunity to add a sensor The electrical measurements are embedded in the inverter, which gives the advantage of not having to purchase and install separate sensors, but the user should consider if the range and accuracy of the measurements are sufficient This is unlikely to be a problem for the small system user, but larger users might want to extend the range of measurements or use duplicate sensors to enhance reliability While the inverter-based systems will give some level of data interpretation, they usually need the user to interpret whether the

PV system is operating according to expectation In some countries, there are also options to purchase a monitoring service to give a higher level of diagnostic capability These services take data directly from the system and compare the values and trends with other systems in the same area This gives a more powerful diagnostic capability than for a single system, since it allows the identification

of systems not operating in line with the majority of other systems

For still larger systems, where most or all of the electricity generated is fed directly into the grid and for which any downtime would result in severe financial penalties, two other measures are often adopted First, the system supplier will offer a combined maintenance and monitoring package, with a guarantee as to the speed with which operational problems will be identified and addressed Second, the supplier is often required to provide a performance guarantee for the output of the system These two measures are clearly linked, since it is advantageous both to the owner and the supplier to ensure that a robust performance monitoring system is in operation

Performance guarantees must take account of the variability of the climatic conditions, which would affect the overall output of the system but are outside the control of the supplier and do not reflect on the quality of the system The most common performance parameters used are a guaranteed minimum final yield or a guaranteed minimum PR In terms of yield, the value needs to be corrected for the irradiation conditions in the period (usually 1 year) compared with the reference conditions used to set