Take a car that loses half a percent of its power every year. After ten years, it will deliver, instead of 100 horsepower, only 95. Even a car enthusiast would hardly notice that reduction. Things are different with photovoltaic arrays, for five percent lower output means around five percent less income, which causes investors a lot of pain. To prevent sudden shocks, yield assessments often anticipate an annual degradation of up to one percent.
Led by Klaus Kiefer, experts at the Fraunhofer Institute for Solar Energy Systems ISE collected data to check on whether this drop in viability actually corresponds to reality, and evaluated it for the last EU PVSEC (Klaus Kiefer, Daniela Dirnberger, Björn Müller, Wolfgang Heydenreich, Anselm Kröger-Vodde: A Degradation Analysis of PV Power Plants, EU PVSEC 2010). The institute now monitors arrays totaling more than 75 megawatts of output. Their study examines 17 systems that have been running for more than five years and together have a nominal power of around ten megawatts.
Statistical analysis requires such a large number of systems in order to reduce the disturbing influence of measurement uncertainty. But anyone who thinks its simply a matter of listing outputs and then extrapolating the output from the arrays is making a mistake. After all, you need precise measurements of insolation on each individual array. If we multiply radiated energy by module efficiency, we get the electrical energy that the array would supply if it worked perfectly and the modules were at a constant temperature of 25 degrees. Scientists express actual yield as a ratio of this value to calculate a performance ratio, and this as a rule lies between 80 and 90 percent. This figure is a measure of how well a system actually uses the locally available insolation, given the efficiency of the modules it uses. If we want to check whether the degradation expressed by the performance ratio is less than one half of a percent annually, thus less than 2.5 percent after five years, we have to measure the performance ratio with corresponding precision. Even a low percentage of measurement uncertainty will make measurements for this purpose useless.
Measuring insolation precisely
One of the most problematic quantities here is the solar radiation continuously recorded by sensors. Lets suppose that the sun shines with the same intensity at specific moments at the beginning and end of the measurement period. If the insolation sensor degrades by, say, four percent during this period, a value four percent lower will appear in the measurement records. If the output from the array has not fallen in this time, the uniform yield will be correlated with an insolation level apparently four percent lower. The array will thus appear to have become four percent better, which is not the case.
To prevent such errors, Kiefer and his colleagues recalibrated almost all the insolation sensors for the arrays every two years. But even calibrated sensors only have a certain precision. The exact insolation values may differ from those measured by up to two percent. This uncertainty alone is as great as the 2.5 percent degradation that the authors want, as far as possible, to exclude from their study. But this two-percent uncertainty isnt all that significant, because the research is not aimed at comparing different systems that have different sensors, but only at determining whether output within a system falls with time. The stability of the sensor is what counts above all.
Repeated calibration showed that, for 70 percent of all the sensors, the discrepancies lay within the measurement uncertainty of two percent (see chart, p. 142). Most of the recalibration values lay between minus one and plus 1.5 percent. The ISE experts used these values to correct the monitoring values continuously. Average discrepancy was only around 0.2 percent. Averaged over many insolation sensors, the authors say this value gives a rough estimate of how stably the sensors work over the period considered.
Unknown temperature
Apart from solar radiation, another significant environmental effect that has to be taken into account is temperature. Temperature sensors were glued to the backs of some modules. Their measurements are subject as a rule to an uncertainty of up to 1.2 percent. These sensors had come unstuck in some older arrays included in the study, so the authors had to correct their measured values, and the uncertainty in the corresponding measured values is greater. The voltage from the solar generator, required to calculate output, and the energy generated on the alternating-current side, can also only be determined with a precision of one to two percent. All these errors add up to a measurement uncertainty of 3.2 to 4.3 percent. The systems are fitted with measuring instruments of differing quality, producing this wide range. The same, however, applies to this total uncertainty as to the uncertainty in the measurement of insolation. It contains systematic fractions significant only if we want to determine the performance ratio in absolute terms. If we use the same measuring technology to measure deterioration in the performance ratio of an array, these fractions are no longer important. Whats important above all is that the measuring instruments remain stable.
A similar argument applies to the problem of the arrays becoming dirty. As long as solar modules and insolation sensors become dirty simultaneously, we can measure the same performance ratio unless the array degrades.
Klaus Kiefer and his colleagues prevented a further measurement uncertainty by only taking account of values measured when the sun was shining at high intensity and insolation sensors measured values of between 800 and 1,000 watts per square meter. This approach automatically excluded values measured when sunshine was incident at a flat angle and uncertainties large. The scientists handled measured temperatures in the same way. Only data for which temperatures fell within a range five degrees wide went into the analysis to prevent errors due to dependence of the performance ratio on the average temperature in individual years.
The result is performance ratio values for each year and each system distributed around a mean value. In one system, for example, they varied over a range of 81 to 83 percent. Using statistical methods, the ISE authors calculated a degradation trend of minus 0.14 percent, or 0.1 percent per year, for this system.
No degradation worth mentioning
When the ISE researchers took all the systems investigated (both monocrystalline and polycrystalline modules), degradation averaged minus 0.1 percent per year (see chart above, p. 142). This shows that it is not in principle necessary to include a system degradation of half a percent in yield assessments, says Klaus Kiefers colleague, Daniela Dirnberger. The three systems using cells made of EFG (edge- defined film-fed growth) silicon, on the other hand, showed a mean degradation of 0.6 percent per year. The researchers consider this a plausible figure, as they had expected such behavior from this cell material. With such systems, therefore, degradation should always be taken into account, says Dirnberger.
Anyone trying to measure degradation values on an individual system shouldnt be alarmed if a greater aging effect shows up. It may even be positive. Remember that the crux of the matter is the analysis of measurement uncertainty. The ISE experts detected variance ranging from minus 1.8 to plus 1.3 percent in the degradation values for systems made from monocrystalline and polycrystalline silicon. This is not surprising, as recalibrating the insolation sensors showed similar discrepancies, and deviation in the measurement of insolation is directly reflected in discrepancies in the degradation analysis. This also explains the apparently positive results of aging, which of course cannot be correct.
Measurement uncertainties average out in the degradation analysis, however, just as in the recalibration of the insolation sensors. Only this fact gives significance to the statement that, in the view of the ISE researchers, aging need not be taken into account in output predictions for monocrystalline and polycrystalline modules.
Measurement uncertainty in the various variables measured | ||
---|---|---|
worst case | best case | |
Measured insolation at 45 degrees Celsius | 2,0% | 2,0% |
Module temperature | 1,2% | 1,2% |
DC current | 0,6% | 0,6% |
DC voltage | 1,6% | 0,3% |
AC energy | 2,0% | 1,0% |
The result is uncertainty in the performance ratio of | 4,3% | 3,2% |
We cant correctly interpret a measurement without knowing measurement uncertainty. The ISE experts therefore estimated the measurement uncertainty at an insolation intensity of 800 watts per square meter and a module temperature of 45 degrees Celsius. Since the measuring instruments on the arrays were different, a value is given in each case for the best and the worst case. Part of the measurement uncertainty is due to systematic measurement errors. If a sensor constantly reports two percent more insolation than actually exists, this measurement error admittedly falsifies the measurement of the performance ratio, but it does not falsify the measurement of the change in the performance ratio over time. Therefore, say the ISE experts, we can use the measurements despite their uncertainty to make an estimate of the actual degradation, which is in the best case 3.2 percent.
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: editors@pv-magazine.com.