Long read: Shadow boxing


From the June edition of pv magazine global

Each year module manufacturers present their R&D teams’ latest achievements – new modules with record-breaking conversion efficiencies. But when inverter manufacturers show off their latest products, it isn’t so often that they include their new devices’ efficiency ratings in the headlines. The reason why is simple. Efficiency ratings for inverters are not entirely straightforward; competition over a tenth of a percentage point on an efficiency rating for an inverter is not particularly noteworthy.

Still, there is a lively exchange of views about system design and technology choices to achieve the highest efficiencies in the field of power electronics. When and how to use module-level power electronics (MLPE) is one such debate. Using MLPE bears a range of value propositions, from string length flexibility, reduced BOS cost, to better monitoring. However, its central function is distributed maximum power point tracking (MPPT).

Initially used in residential systems to handle non-uniform shading from trees or chimneys, DC power optimizers are now being rolled out in the gigawatt range to commercial and industrial systems, as well as ground-mounted arrays. It’s a trend that can be traced back to the United States, after the U.S. National Electrical Code was amended to require rapid shutdown devices on all rooftop installations. MLPE vendors seized the opportunity and suggested that DC optimizers and monitoring devices be installed, with rapid shutdown capability.

However, it is not only in the United States that sales numbers for MLPE systems have been steadily climbing. What is clear is that partial shading events are less likely to occur on C&I rooftops or in ground-mount installations. So, if the argument for MLPE was that it handled partial shading better, how do such devices stand up against string installations in full sun?

That question was what informed a report by Toke Franke titled “The Impact of Optimizers for PV-Modules,” which examined a 42-module test setup, with and without DC optimizers. Franke’s conclusion was that MLPE in most scenarios lead to lowered yield compared to ‘modern string inverters’. The research was disseminated by SMA and discussed throughout the industry. SolarEdge, meanwhile pointed to a series of methodological weaknesses of the study, such as handpicking of data, to rebut the claims that were made.

Finding maxima

When a PV module produces power, its surrounding conditions will alter its maximum power point (MPP). As power equates to voltage times current, there is an ideal ratio between the two dubbed MPP, for each condition. On a hotter day, the modules and the string will produce less voltage, while on a colder day, the voltage will rise. Changing irradiance by partial shading, on the other hand, will alter current.

If conditions for all modules are precisely the same, there is a single MPP for all modules in a string. Non-uniform shading from bird droppings, obstructive objects such as poles or chimneys, incorrect azimuths, non-linear degradation, and otherwise mismatched modules create multiple local peaks. “So, all inverters should track for the global rather than the local maximum. Not every manufacturer manages to develop an MPPT algorithm that does that very well,” explains Jenya Meydbray, CEO of PV Evolution Labs.

Another way of allowing the modules to operate at their highest efficiency is to use DC optimizers. They run each module at its maximum power point and convert DC to DC to some other voltage configuration that the string inverter can handle best. The flip side is that DC optimizers come at a cost, must be installed on the module, and operate, like all electronics, at less than 100% efficiency.

“From a power electronics point of view, I would say first of all that with the right MPP, more yield per module is achieved,” says Marco Jung, professor for e-mobility and electric infrastructure at the Bonn-Rhein-Sieg University of Applied Sciences, Germany and department head for converter and electric drive technology at Fraunhofer IEE. “However, the question is whether overall the MLPEs don’t nullify the additional yield due to a poorer efficiency of the optimizer compared to a string inverter.”

Efficiency of what?

Deploying a centralized solution is essentially a more efficient circuit. “In my experience, MLPE shows a lower circuit efficiency,” says Jung. “Each power electronics component has its semiconductors with drivers and a range of other components such as relays, mostly a galvanic isolation and microcontrollers, all of which consume power and produce losses.” The components which are indirectly necessary for DC conversion are dubbed “periphery” – displaying relatively constant electrical consumption.

Efficiency ratings on the datasheets of inverters, and DC optimizers, are only of limited help when determining the most efficient solution. Power electronics, both string inverters and MLPE, have efficiency curves as a function of DC power, or voltage, or temperature. The efficiency rating printed on data sheets typically comprises two numbers. One is the maximum efficiency, as a function of power, while the other is the weighted efficiency, as a function of power. One very popular rating is the California Energy Commission (CEC) weighted efficiency. “The CEC efficiency rating very heavily weights the amount of power at 75%. So, if it is a 10 kW inverter at 7.5 kW of input power is the most heavily weighted power level in the curve,” Meydbray says.

Weighting efficiencies

Euro Efficiency = 0.03 x Eff5% + 0.06 x Eff10% + 0.13 x Eff20% + 0.1 x Eff30% + 0.48 x Eff50% + 0.2 x Eff100%.

CEC Efficiency = 0.04 x Eff10% + 0.05 x Eff20% + 0.12 x Eff30% + 0.21 x Eff50% + 0.53 x Eff75%.+ 0.05 x Eff100%.

When the CEC weights its efficiency heavily towards 75%, it does so because it expects systems in arid regions of high insolation to operate at 75% most of the time. Weighting the power input makes sense, but only regionally. The time the inverter will spend at 75% of its power rating is different in California than in Scandinavia, for example. Solar arrays in Scandinavia achieve between 650 and 800 full load hours per year. In southern Europe, it can be around 1,500. The assumption in Europe is that the array will operate at 50% of its rated power most of the time. Therefore, the European weighting method takes much more into account the efficiency at 50% power, to better reflect the real-world conditions in Europe. In the case of either weighting method, both inverter and MLPE operate at less than 100% efficiency. The efficiency curve as a function of temperature, voltage, or power might be a different shape for a DC optimizer than it is for an inverter.

When the device adjusts the power in the DC to DC converter stage, some energy is lost. Depending on the manufacturer of the device and surrounding conditions, something between 0.7% and 4% will be lost. Additionally, the DC to AC conversion stage will lose another 1% to 4% to thermodynamics, depending on the manufacturer, but this also depends to a substantial degree on the input power and surrounding conditions.

This conditionality of efficiency is crucial to drawing conclusions on using DC optimizers. The efficiency curves have different shapes for each inverter and DC optimizer. That is the case even when the efficiency rating of the two products is the same. An inverter and an optimizer can have a Euro-weighted efficiency of 97%, but their efficiency at 80% DC input power might be different. This means that for each condition (temperature, input power, voltage), the exact efficiency rating of both the inverter and optimizer must be known.

“There are definitely conditions where power optimizers will perform better than a single central inverter, and there are definitely conditions where the opposite is true,” Meydbray explains. He adds that DC optimizers can outperform central inverters in yield even in completely unshaded conditions when surrounding factors line up. “So, both SMA and SolarEdge can find real studies that say that their technology is better, and those studies are definitely true.” Meydbray adds that the results of these studies don’t translate across the board and it isn’t possible to draw any conclusions of universal character.

Both Meydbray and Jung struggle to give a rule of thumb about when to use MLPE and when not. The PV researchers point to the nuances and complexities of real-world applications, where many factors occur simultaneously. For example, ambient temperature affects efficiency and specific voltage window. Inverters cannot operate at the highest voltage configuration and the highest ambient temperature simultaneously without derating – even if the manufacturer rightfully stipulates that the device will not start to de-rate at that particular temperature. While it may be difficult to pinpoint each condition in which MLPEs are best employed, there is consensus that the more non-uniform the shading on the array, the better the fit for MLPE. “For perfectly well-irradiated situations with zero shadings, typically MLPE won’t be worth the added cost,” concludes Meydbray.

Quizzed about a potential test sequence that could provide credible data to inform whether a DC optimized setup will provide more yield than a regular string inverter setup, Jung suggests an emulation. A test sequence that seeks to determine the efficiency of the power electronics arrangement would need to exclude any inaccuracies due to solar modules coming mismatched from the factories, for example. “For real comparability of the power converter systems, the conditions must be the same,” says Jung. “In a ‘power hardware in the loop’ environment, the PV module’s characteristic curve, behavior, ambient conditions, and the grid connection should be realistically reproduced. In this way, a reproducible test and comparison of the converter systems can be carried out, and errors can be reproducibly mapped.”