Calibration of photovoltaic reference cells by the global sunlight method

INSTITUTE OF PHYSICS PUBLISHING Metrologia 42 (2005) 360 367 METROLOGIA doi:10.1088/0026-1394/42/5/004 Calibration of photovoltaic reference cells by the global sunlight method Harald Müllejans, Willem Zaaiman, Ewan D Dunlop and Heinz A Ossenbrink European Commission, Directorate General Joint Research Centre, Institute for Environment and Sustainability, Renewable Energies Unit, European Solar Test Installation, TP 450, Via Fermi 1, I-21020 Ispra (VA), Italy E-mail: harald.muellejans@cec.eu.int Received 1 December 2004 Published 1 August 2005 Online at stacks.iop.org/met/42/360 Abstract The global sunlight method for the calibration of reference photovoltaic cells is described and illustrated with results from recent measurements. In this method, the short circuit current of the solar cell is recorded as it tracks the sun s position. This current, corrected for temperature and total irradiance deviation, is then plotted against pressure-corrected geometric air mass (AM). The calibration value (CV) at standard test conditions (25 C, 1000 W m 2 total irradiance with AM 1.5 spectral distribution) is extracted by a straight line fit to the data. The global sunlight method is less complex than other methods as it avoids (in particular conditions) the need for spectral irradiance measurements and corresponding relative spectral response determination. It has an estimated expanded uncertainty of 1% and agrees to better than 0.5% with the CVs of the World Photovoltaic Scale. (Some figures in this article are in colour only in the electronic version) 1. Introduction In the rapidly developing photovoltaic industry, the determination of the electrical performance of photovoltaic devices is assuming increasing importance. The traceability chain leads from radiometric standards to photovoltaic products via primary photovoltaic reference cells, working and production standards. They can be readily calibrated using solar simulators as all standards from the primary reference cell downward are photovoltaic devices. These calibration methods are well established and are shown to have high accuracy and repeatability. The largest uncertainty in this transfer arises from the calibration of the primary reference cell. The reference cells produce a current proportional to the total irradiance; this current depends linearly on the irradiance, close to standard test conditions (STC) which define a device temperature of 25 C, a total irradiance of 1000 W m 2 and a spectral irradiance distribution according to [1]. In the irradiance range of interest (around 1000 W m 2 ) the reference cells can be calibrated against the world radiometric reference (WRR) [2]. The WRR has been shown to be identical to SI units for all practical purposes [3]. This method, however, requires a continuous light source with sufficient intensity (a continuous solar simulator or natural sunlight). The reference solar cell is an irradiance sensor, like a pyrheliometer or a pyranometer, with the distinctive advantage that it has a much faster response (order of microseconds) so that it can also be used on pulsed solar simulators. While pyrheliometers and pyranometers are readily calibrated traceable to WRR, this is not the case with reference solar cells. In fact, for the former there are several international standards [4], whereas for the latter there is only a draft (PWI: 60904-4 in [5]). For the most common crystalline silicon (c-si) solar cells, calibration may be performed under simulated continuous sunlight [6] (requiring the determination of the absolute irradiance of the light source, its spectral distribution and the relative spectral responsivity of the reference solar cell. Alternatively, the calibration may be made by absolute spectral responsivity [7, 8]. A third option is to calibrate with natural sunlight. For calibration with natural sunlight, there are two options available, the first being the calibration against direct sunlight, 0026-1394/05/050360+08$30.00 2005 BIPM and IOP Publishing Ltd Printed in the UK 360

Calibration of photovoltaic reference cells by the global sunlight method excluding the diffuse component by a suitable collimator [9 11]. This option has the advantage of being able to use a pyrheliometer to measure the direct solar irradiance; however, it requires the spectral irradiance distribution over a large wavelength interval (300 nm to 4000 nm) and the relative spectral responsivity of the reference cell. The second option is to use global sunlight. This adds some complexity to the measurement of the irradiance as the total irradiance is the sum of direct (by pyrheliometer) and diffuse (by shaded pyranometer) irradiance; however, it has the enormous advantage of making the measurement of the spectral irradiance and the relative spectral responsivity of the reference cell obsolete. This arises from the fact that natural global sunlight matches the defined spectral irradiance distribution more closely than any simulated light. The defined spectral irradiance distribution was originally calculated for a geometric air mass (AM) of 1.5 with certain atmospheric parameters. Therefore, the defined spectral irradiance distribution can be approximated by performing the measurements at similar atmospheric conditions. To reduce noise and natural variability, the measurements are performed against a range of AM values around AM = 1.5 and the calibration value (CV) is determined from a straight line fit at AM = 1.5. The other two conditions (temperature and total irradiance) are readily achieved or corrected for. The match of spectral irradiance distribution is critical owing to the large difference in spectral responsivity, c-si being limited to the wavelength interval 300 nm to 1200 nm, whereas a pyrheliometer typically detects up to 4000 nm. A correction for this spectral mismatch is mathematically possible [12], but requires the knowledge of both the spectral irradiance distribution of the light source and the spectral responsivity of the reference cell. The spectral mismatch vanishes when the spectral irradiance distribution approaches that of the standard [1] such that the associated uncertainty in determining the spectral mismatch becomes larger than, or of the same order as, the actual correction. The global sunlight method makes use of this fact by assuming that this is the case for natural global sunlight at AM = 1.5. The discrepancy remaining between natural sunlight and the standard is accounted for in the uncertainty budget, and reduced through repeated measurements on three different days. 2. Global sunlight method The measurements were performed at the European Solar Test Installation (ESTI) at Ispra in northern Italy (45 48 43.4 North 8 37 37.4 East, 220 m altitude). 2.1. Reference cells and instrumentation Six c-si reference cells were calibrated, four of which (PX102C, PX201C, 930417-1 and 930417-2) form part of the World Photovoltaic Scale (WPVS) [13] and two new cells (PX301C and PX304C) in the standard PRC Krochmann package, which are the same as PX201C, with the exception of the internal temperature sensor which is a PT100 on the PX3xxC series. The direct solar irradiance was measured with two practical absolute cavity radiometers (PACRAD) (PMO6 81109 and PMO6 911204), both calibrated at the International Figure 1. View of a typical measurement set-up with the reference cells on the left, the shaded pyranometer for diffuse irradiance in the centre, and the two cavities (square shaped) for direct irradiance on the right. There are also two pyrheliometers (round shaped) mounted on the right. Pyrheliometer Comparison IPC IX in Davos, Switzerland in 2000 [14]. A reference pyranometer was calibrated against these PACRADs by the continuous sun and shade method [4]. For the reference cell calibration, the pyranometer was used with a shading disc subtending the same angle as the aperture angle of the cavities. Furthermore, a solar tracker, temperature controllers for the reference cells and the mounting plate of the reference cells, internal temperature monitor of reference cells, and current-to-voltage converters were used. All data were recorded with an HP34970A data logger. The air pressure was read from a pressure meter. 2.2. Measurement procedure Only days with clear sky conditions were selected. The reference cells and the irradiance sensors (2 PACRADs and shaded pyranometer) were mounted co-planar on a sun tracking platform. A typical set-up (figure 1) shows the two PACRADs on the right, the shaded pyranometer in the centre and the reference cells on the left. During this particular measurement, two additional pyrheliometers were also mounted for other purposes. The temperature of the reference cells was maintained at (25 ± 2) C by water cooling applied directly to the reference cell (where possible, i.e. 930417-1 and 930417-2) or to the mounting plate. The signals of the two cavities (direct irradiance), the shaded pyranometer (diffuse irradiance), the internal temperature sensors of the reference cells and their short circuit current (via current-to-voltage converter) were read at intervals of 80 s (as the cavities need two time periods of 40 s for each reading). Typically, the measurements were performed from 07:00 to 17:00 (UTC + 1 h) while tracking the Metrologia, 42 (2005) 360 367 361

HMüllejans et al Table 1. CV obtained by the global sunlight method in spring 2004. I sc /ma Date Period PX102C PX201C 930417-1 930417-2 PX301C PX304C Average 28 March 2004 Morning 115.80 122.87 124.76 123.27 124.55 123.85 29 March 2004 Afternoon 116.48 122.90 125.12 123.38 124.52 123.87 01 April 2004 All day 117.19 123.65 125.65 124.04 125.27 124.59 CV Average 116.49 123.14 125.18 123.57 124.78 124.10 Standard deviation 0.69 0.44 0.45 0.42 0.43 0.43 Standard deviation (as a percentage) 0.59% 0.36% 0.36% 0.34% 0.34% 0.34% 0.39% sun s position with a precision of better than 0.1, solar noon being around 12:30. The geometric AM was calculated for the calibration location (including altitude correction) assuming standard air pressure, and corrected for actual air pressure by multiplying with the ratio of measured to standard air pressure. The measurements were performed for pressure-corrected AM from AM 3 to AM below 1.5, and back to AM 3. For the Ispra location, minimum AM below 1.5 occurs only on days in the period from mid-march to the end of September. In order to be able to measure up to AM 3 (considering sufficient data points and instrument warm-up time) the period for Ispra is from the beginning of September to mid-april of the following year. Both conditions are fulfilled in the two overlaps of the two time periods, i.e. in spring (mid-march to mid-april) and in autumn (September). Here, we report the results of the spring calibration campaign in 2004. Measurements were performed on 28 and 29 March and 1 April 2004. 2.3. Data analysis The acquired data were filtered and analysed according to the following criteria. Reject data points where clouds (which could develop during the day) were present either in front of the sun or close to it. This is easily identifiable from the plot of the direct irradiance against the time of the day. Reject data points where the readings of the two cavities differed by more than 0.2%. Reject data points which differed from the previous readings by more than 3%. Calculate the direct irradiance H dir as the average of both cavities. Calculate the diffuse irradiance H dif from the pyranometer. Calculate the total irradiance as the sum of both: H tot = H dir + H dif. Reject data points with H tot < 800 W m 2 (or H tot > 1200 W m 2, though this did not occur). Make a linear correction of the short circuit current of all reference cells to 1000 W m 2 irradiance. Correct the short circuit current to a temperature of 25 C with the temperature coefficient α of the reference cell. This step is optional, and required only if the temperature was outside the interval of (25 ± 2) C, as otherwise the uncertainty is almost negligible (see below). Calculate the diffuse ratio = H dif /H tot and plot against time and pressure-corrected AM. Remove data points with a diffuse ratio below 10% or above 30%. Plot the short circuit current of the reference cells against pressure-corrected AM. Delete obviously erroneous readings. The data points are spaced unevenly for different AM: for low AM (around solar noon) there are several readings for similar AM, whereas for high AM (morning and afternoon) there are fewer readings. In order to balance the fit (see below), all valid short circuit current readings were averaged into 0.01 bins in AM. A straight line was fitted to the binned short circuit currents against pressure-corrected AM in the range where the data are linear, typically below AM 2, and for AMs somewhat larger than the minimum AM occurring at solar noon. The CV was calculated from the straight line fit equation for AM = 1.5. The fit was made to the combined data of the morning and afternoon. Some unexpected changes in the weather resulted in only one of the two branches being made available for two out of the three days. The final CV was taken as the average of all three days. 3. Results Typical results are presented in graphical form and all final results are summarized in table 1. For 28 March, only the morning data were considered, as cloudy conditions prevailed around the time when AM = 1.5 was attained in the afternoon. For 29 March, only the afternoon data were considered, again because cloudy weather prevailed until AM = 1.5 was attained in the morning. For 1 April, although cloudy conditions existed around solar noon, the sky was clear around AM = 1.5 both in the morning and in the afternoon, and therefore the data for the whole day were considered. The various irradiances and the diffuse to total ratio for 1 April 2004 are presented (figure 2). The gap in the data above AM2 for the morning branch is due to a stoppage of the data acquisition program. As an illustration, the data and the fit for the two WPVS cells (PX102C and PX201C) are shown for 28 March (figure 3) and 29 March (figure 4). It may be noted that the slope of the fit is negative in the morning and positive in the afternoon, which was confirmed on all cells and for all days. The data for 1 April (PX201C) show both branches (figure 5). The fit for the morning yields 123.33 ma whereas that for the afternoon yields 124.02 ma. The fit for the entire day (not shown) yields 123.65 ma, which is close to the average of the two values. This shows that it is acceptable to combine the data 362 Metrologia, 42 (2005) 360 367

Calibration of photovoltaic reference cells by the global sunlight method Irradiance [W/m 2 ] Hdif / Htot Figure 2. Irradiance (direct, diffuse and total) as well as diffuse to total ratio plotted against pressure-corrected AM for 1 April 2004. 124 123 Isc [ma] @1000W/m 2 @ 25C 122 121 120 119 118 117 116 y = -1.4598x + 125.06 PX102C PX201C Linear (PX102C) Linear (PX201C) y = -0.8057x + 117.01 115 114 1.4 1.5 1.6 1.7 1.8 1.9 2.0 Pressure corrected AM Figure 3. Short circuit current data and linear fit versus pressure-corrected AM for PX102C and PX201C taken on 28 March 2004 (morning). for the entire day rather than treating morning and afternoon separately. The fits were restricted to AM < 2, as the above data did not fit a straight line or were sparse, for instance on 1 April for the morning branch, because the ratio of diffuse to total irradiance was above 30%. An unexpected drop in the measured short circuit current was observed for PX304C in the afternoon branch of 1 April Metrologia, 42 (2005) 360 367 363

HMüllejans et al 126 125 124 Isc [ma] @1000W/m 2 @ 25C 123 122 121 120 119 118 117 116 y = 1.6988x + 120.35 PX102C PX201C Linear (PX102C) Linear (PX201C) y = 1.7312x + 113.88 115 114 1.3 1.4 1.5 1.6 1.7 1.8 1..0 Pressure corrected AM Figure 4. Short circuit current data and linear fit versus pressure-corrected AM for PX102C and PX201C taken on 29 March 2004 (afternoon). Isc [ma] @1000W/m 2 @ 25C 130 129 128 127 126 125 124 123 PX201C 01/04/2004 pm PX201C 01/04/2004 am Linear (PX201C 01/04/2004 pm) Linear (PX201C 01/04/2004 am) y = 4.4891x + 117.29 y = -1.2995x + 125.28 122 121 120 1.3 1.4 1.5 1.6 1.7 1.8 1..0 Pressure corrected AM Figure 5. Short circuit current of reference cell PX201C taken on 1 April 2004. The measurements for the morning (am) and the afternoon (pm) are plotted and fitted separately. 364 Metrologia, 42 (2005) 360 367

Calibration of photovoltaic reference cells by the global sunlight method 127.0 126.5 PX304C 01.04.2004 afternoon 126.0 Isc [ma] @1000W/m 2 @ 25C 125.5 125.0 124.5 124.0 123.5 123.0 1.3 1.4 1.5 1.6 1.7 1.8 1..0 Pressure corrected AM Figure 6. The measured short circuit current of PX304C taken on 1 April 2004 (afternoon) shows a drop for AM > 1.7, which is not explained. The fit result for AM = 1.5 is insensitive to whether these data points are included or not. (figure 6). The encircled points are lower than expected and would normally be removed as erroneous measurements. However, the fit gives nearly the same result, whether these points are taken into account or omitted. This shows that the fitting makes the result insensitive to erroneous readings, as long as these are not close to AM = 1.5. The standard deviation for all cells is 0.36% or less (table 1), with the exception of PX102C which gives 0.59%. On average, the standard deviation for all six cells is less than 0.4%. Taking the average of the maximum deviations from the average yields similar values. 4. Uncertainty analysis The short circuit current I sc is calculated as I meas 1000 W m 2 I sc =, (1) 1+α(T 25 C) H tot where I meas is the measured short circuit current of the reference cell, H tot = H dir + H dif the total irradiance, α the relative temperature coefficient for short circuit current and T is the temperature of the reference cell. The uncertainty was calculated considering all the components (table 2) as single standard deviation (k = 1). The electrical uncertainty of the short circuit current measurement contains a contribution from the current-to-voltage converter (0.058%) and the data logger (0.029%). The maximum temperature deviation from 25 C is ±2 C. At a typical temperature coefficient for c-si of 500 ppm C 1 this yields 2 C 500 ppm C 1 = 0.10%. The direct irradiance is measured by two absolute cavity radiometers calibrated Table 2. Uncertainty budget for the global sunlight method. Component % Electrical sqrt (0.058 2 +0.029 2 ) 0.07 Temperature ±2 C 500 ppm C 1 0.10 Irradiance H tot = H dir + H dif 0.22 Spectral Deviation 0.20 irradiance from IEC60904-3 : 1989 distribution Combined k = 1 0.32 standard uncertainty Expanded k = 2 0.64 uncertainty at IPC-IX in Davos (2000) [14], whose standard deviations are 0.06% and 0.08%, i.e. on average 0.07%. The data logger as well as the offset of the WRR to SI units each contribute 0.029% [3]. This gives a combined uncertainty for the direct irradiance of 0.08%. The diffuse irradiance measured by the shaded reference pyranometer has a 0.78% uncertainty (from the calibration of the pyranometer against the PACRADs). The total irradiance is the sum of direct and diffuse components. Typically, it is composed of 80% direct and 20% diffuse irradiance. The uncertainty is therefore 0.8 0.08% + 0.2 0.78% = 0.22%. The uncertainty arising from the spectral irradiance occurring in the natural sunlight as opposed to that defined in IEC 60904-3 : 1989 is estimated to be 0.2%, based on the results in [18]. The uncertainty in AM can be neglected, as a 0.5% (about 5 mbar) uncertainty in pressure leads to a 0.5% uncertainty in AM, which is less than 0.01 at AM < 2. Metrologia, 42 (2005) 360 367 365

HMüllejans et al Table 3. Comparison of the CVs in milliamps determined by the global sunlight method with the values of the WPVS [16]. PX102C PX201C 930417-1 930417-2 WPVS (1998) 116.76 123.29 124.87 122.97 CV (average) 116.49 123.14 125.18 123.57 Difference 0.23% 0.12% 0.25% 0.48% to WPVS The standard deviation of the values determined on different days has to be added to the uncertainty reported in table 2. For the data reported here, the daily variation was 0.39% (k = 1) (table 1). This leads to a final overall combined standard uncertainty for the global sunlight method of 0.50% (k = 1), and an expanded combined uncertainty of 1.00% (k = 2). 5. Discussion From the results it is expected that the global sunlight method determines the CVs of c-si reference solar cells with an expanded uncertainty of ±1%. Four of the cells measured here are part of the reference cell set for the WPVS, which were originally calibrated in 1994 1996 [15] and recalibrated in 1998 [16]. In table 3, the WPVS values of the four cells are compared with the average obtained by the global sunlight method and the largest deviation is found to be less than 0.5%. This confirms that the calibration by the global sunlight method is valid, and that the uncertainty budget is consistent, especially when considering that the original WPVS calibration had an expanded combined uncertainty of ±1.9% [13]. This large uncertainty was obtained by averaging the results of four different laboratories, all using different calibration methods (none of which used the global sunlight method). As it was not possible at the time to determine which among the various methods was more reliable, the results have been averaged. Based on the results presented here, the global sunlight method has been shown to be an alternative method capable of delivering CVs of the same quality as the other methods. The results presented here confirm that the global sunlight method is a traceable and reliable method for solar cell calibration. In the global sunlight method, a difference between morning and afternoon measurements is observed here, which is most prominent in the slope of the fit which is negative in the morning and positive in the afternoon (figures 3 5). Several parameters recorded on the meteo tower [17] situated next to the test site were investigated for systematic behaviour in am/pm conditions but no clear correlation was found. One possibility could be the aerosol content, but this parameter was not measured. The deviations between the morning and afternoon measurements became greater in the summer [18], with particularly large deviations ocurring in the afternoon. Work is in progress to verify the global sunlight method in other geographic locations. The idea of determining the CV at AM = 1.5 from a fit to the data is based on the Langley-plot, which is used to determine the CV at AM = 0 (for space cells) from measurements at low AMs during high altitude aircraft flights [19]. It should be noted, however, that in a Langley-plot the logarithm of the short circuit current is plotted against AM and that it is extrapolated to AM = 0. The global sunlight method is (relatively) easy to use, and does not require the determination of either the relative spectral responsivity of the solar cell to be calibrated or of the relative spectral irradiance distribution during calibration. The latter is particularly cumbersome, as many commercially available spectroradiometers measure over a limited range of wavelength, which additionally requires then the modelling of the spectrum outside the measured wavelength range. The global sunlight method avoids not only these additional measurements, but also the associated problems of traceability and their complex contributions to the uncertainty budget. The global sunlight method also has no major restrictions with regard to the photovoltaic device size. The greatest difficulty encountered in the global sunlight method is in obtaining sufficient days with good, stable weather conditions. 6. Conclusions The global sunlight method has been applied to six c-si reference cells, four of which are part of the WPVS reference set. From the uncertainty budget calculation and from the comparison with previous WPVS CVs, it can be concluded that by using the global sunlight method it is possible to deliver CVs for reference solar cells with the same reliability offered by other established methods. The main advantage of the global sunlight method is the reduced complexity, leading also to a reduction in the investment required for the instrumentation. Acknowledgment We thank the Joint Research Centre for support within its institutional programme (Exploratory research in Solar Electricity Action 2324). References [1] IEC 60904-3 1989 Measurement principles for terrestrial photovoltaic (PV) solar devices with reference spectral irradiance data [2] Fröhlich C 1991 History of solar radiometry and the world radiometric reference Metrologia 28 111 15 [3] Romero J, Fox N P and Fröhlich C 1995/96 Improved comparison of the World Radiometric Reference and the SI radiometric scale Metrologia 32 523 4 [4] ISO 9059 1990 Solar energy calibration of field pyrheliometers by comparison to a reference pyrheliometer ISO 9846 1993 Solar energy calibration of a pyranometer using a pyrheliometer ISO 9847 1992 Solar energy calibration of field pyranometers by comparison to a reference pyranometer [5] Ossenbrink H 2003 Calibration procedures state of the art Proc. 3rd WCPEC (Osaka, Japan, 2003) pp 2177 81 [6] Shimokawa R, Nagamine F, Miyake Y, Fujisawa K and Hamakawa Y 1987 Japanese indoor calibration method for the reference solar cell and comparison with outdoor calibration Japan. J. Appl. Phys. 26 86 91 [7] Metzdorf J 1987 Calibration of solar cells. 1: The differential spectral responsivity method Appl. Opt. 26 1701 8 366 Metrologia, 42 (2005) 360 367

Calibration of photovoltaic reference cells by the global sunlight method [8] Metzdorf J, Winter S and Wittchen T 2000 Radiometry in photovoltaics: calibration of reference solar cells and evaluation of reference values Metrologia 37 573 8 [9] Osterwald C R, Emery K A, Myers D R and Hart R E 1990 Primary reference cell calibrations at SERI: History and methods Proc. 21st IEEE PVSC (Kissimmee, FL, 21 25 May 1990) pp 1062 7 [10] ASTM E 1125-99 1999 Standard test method for calibration of primary non-concentrator terrestrial photovoltaic reference cells using a tabular spectrum [11] Bücher K, Stiening R, Heidler R, Emery K, Field H, King D and Hansen B 1993 Intercomparison of two primary reference cell calibration methods Proc. 23rd IEEE PVSC (Louisville, KY, 10 14 May 1993) pp 1188 93 [12] IEC 60904-7 1998 Computation of spectral mismatch error introduced in the testing of a photovoltaic device [13] Osterwald C R et al 1999 The world photovoltaic scale: an international reference cell calibration program Prog. Photovoltaics 7 287 97 [14] 2001 International Pyrheliometer Comparison IPC-IX 2000 Results and Symposium Working Report No 197, Meteo Swiss [15] Osterwald C R et al 1998 The Results of the PEP 93 Intercomparison of Reference Cell Calibrations and Newer Technology Performance Measurements: Final Report NREL/TP-520-23477, p 209 [16] Emery K 2000 The Results of the First World Photovoltaic Scale Recalibration NREL/TP-520-27942, p 14 [17] http://sunbird.jrc.it/meteo/meteo.php [18] Müllejans H, Ioannides A, Kenny R, Zaaiman W, Ossenbrink H A and Dunlop E D 2005 Spectral mismatch in calibration of photovoltaic reference devices by global sunlight method Meas. Sci. Technol. 16 1250 4 [19] Jenkins P, Brinker D and Scheiman D 1997 Uncertainty analysis of high altitude aircraft air mass zero solar cell calibration Proc. 26th IEEE PVSC (Anaheim, USA, 30 September 3 October 1997) pp 857 60 Metrologia, 42 (2005) 360 367 367