In my previous post, I had analysed the economics of my small (6-module) PV system. Somewhat to my surprise, it turns out to be uneconomic, mostly due to it producing far far less than the 2300kWh per annum its installer had estimated it’d produce.
In an effort to understand why the estimate was so inaccurate, I ran my own simulations — using hourly weather data for my location (courtesy of Solcast), and a highly detailed simulator (courtesy of PVsyst). As indicated in my previous post, the six modules on my rooftop were not the Peimar 300s specified in the build list for the installation contract. Instead they are six Trina 270; so the installed PV system has the 1620Wp of photovoltaic panel that is specified in the contract. The installer’s estimated annual yield is reasonable for a 1620Wp system in my location, as 2300kWh/1620Wp = 1.4 kWh/kWp is a readily-achievable efficiency here in sunny temperate Auckland NZ.
I won’t attempt to enumerate all of the blind alleys I explored before I *finally* figured out why the efficiency of my system is below 1.2 kWh/kWp.
The root cause is the variable-cloudiness of Auckland’s skies, which (in combination with a rather slow inverter (Goodwe 3000 NS, vintage 2013) and a somewhat underpowered system (1620Wp of module on a 3000W inverter) spells TROUBLE for the MPPT heuristics.
Warning: only solar geeks will understand any of my abbreviated explanation below.
The ShadowScan feature on my 3000 NS has a 90-minute period. That’s a *long* time to wait for it to recover after getting “stuck” on a bad point (or “going around in circles”) in its MPPT search space. And… sometimes it’s not necessary to wait 90 minutes, because every time a cloud passes by it essentially throws the MPPT into another spot in its search space … and sometimes that will allow the MPPT to “find its way” to a near-optimal operating point.
A not-uncommon “circle” is when the DC voltage bounces (ineffectually!) between the inverter’s 70V minimum and my string’s (roughly) 200V maximum. Alternatively it can get “stuck” in a Wait state at the 200V+ open-circuit voltage.
Such “lost in space” behaviour of the MPPT in my Goodwe 3000NS inverter hugely reduces my PV power in late afternoons.
Below you’ll find a plot of my simulation v actuals for a late Spring day. Note the Wait states at high DC voltage. Peak power on my system is always in the 150V to 170V range, except during an hour in late afternoon in April, when a tī kōuka (aka “cabbage tree”) briefly shades one module. So: any time that the voltage is outside this range, the MPPT hasn’t “found” the sweet spot. How hard can that be, you might ask? Yeah well what’s really required, I reckon, is an inverter with far more computational power than my Goodwe 3000 NS — so that it can complete a “perturb and observe” cycle in a second or three, rather than (as appears to be the case) adjusting the load on the panels only once or twice per minute.
Below are plots from another sample day: in late Spring of last year. The first circled set of voltages are producing power at a greatly suboptimal rate, I suspect because variable cloudiness has caused the MPPT to run at an operating point which would be appropriate only if one of a module’s substrings were shaded — and this doesn’t happen on my rooftop in November.
Finally here’s a comparison between simulation & actuals for a late-winter day. This is a difficult case for any inverter, as the irradiance is low *and* variable. Note that sunlight sufficient to generate 1kW of power (if the MPPT were ideal *and* there was no power lost between the panels and the AC output of the inverter) is at only 1/3 of the peak rated power (3kW) of a Goodwe inverter. I suspect that, at such low levels of irradiance, the minimum “granularity” of its load perturbations is too large (possibly 75W?) to allow its MPPT to find a stable (i.e. near-optimal) operating point.
I have a vague hope that Goodwe has produced an updated firmware for the 3000NS that’s 1) suitable for use in NZ, 2) either has a different set of MPPT heuristics or a different periodicity on its Shadow Scan, and 3) they’re willing to release it to me. Re #1: Goodwe did produce a new version for their Australian market.
Wish me luck!
I reckon I can afford to “pay” roughly 8Wh * 24 = 200Wh per day (for an MPPT scan every 32 minutes), and I’d still come out “ahead” on my current status — of losing about 1kWh per day to an MPPT that is either “stuck” or “going around in circles”.
I believe the Goodwe 3000 NS was commonly deployed in small rooftop PV systems in NZ, so getting a “firmware fix” to its MPPT problems in my variable-cloudy location could be important to many others.
cthombor 