How is treated the gain of string inverters with respect to centralized ones?

[André Mermoud]

It is a usual belief to think that systems with string inverters have a better yield than centralized ones.

I am not sure of this, if there is a differece, it is probably very low.

How can this be taken into account in PVsyst ?

- The main implication I can identify is the electrical mismatch losses under partial shadings.

This graph shows that with one only string on one MPPT,the loss is limited (blue line); but as soon as you have two strings or more in parallel on one MPPT, the electrical yield for beam becomes null as soon as you have 1/3 of the submodules shaded (i.e. the case of sheds, with the bottom cell's row shaded).

See also How to evaluate the effect of by-pass diodes in shaded arrays?.

Electrical Shading loss acc. to number of strings on the MPPT

This diagram shows that the configuration of "One only string per MPPT" will indeed have a lower electrical loss.

Equivalently, if you put all bottom strings of different tables on a same MPPT input, you will get the same result. This configuration should be preferred with string inverters with few MPPT inputs.

- The wire length variations - and therefore wiring voltage drops - are obviously higher for centralized systems. A new tool of PVsyst will show that the mismatch between strings of different voltages is extremely low. As an example: for a 500 kW array and 4mm² string wires, one central inverter would have 0.17% mismatch loss, when distributed string inverters with 2 strings per MPPT would have 0.05%.

- The number of MPPT (centralized or distributed systems) may have consequences on the wiring layout. It is your job to define and optimize the wiring and therefore to calculate the wiring resistance of the PV array or the full system.

- Big inverters may have a higher MPP tracking inefficiency (due to response time). PVsyst doesn't define this inefficiency: it assumes that it is included in the inverter's efficiency profile. However very few manufacturers give information about this inefficiency.

- With big systems we can imagine a mismatch effect because of inhomogeneities due to clouds passages. These are transient phenomenons, not really significant on the hourly accumulations of the simulation. By the way the mismatch loss between strings of inequal irradiances is usually very low.

- Finally, an inverter failure will be easily managed with string inverters, when it penalizese the running of the full centralized systems. This could be taken into account when specifying unavailability losses.

Besides these arguments, I can't see any significant difference between centralized and distributed systems, which could be taken into account by the simulation.