AC-coupled PV inverter total yield reported 1000x too low via SunSpec (AEG AS-IC02-4000-2)

Anyone else had their AC-coupled inverter reporting comically tiny yield figures via SunSpec / Modbus TCP into a Cerbo GX?

Mine was showing lifetime generation that looked like I'd powered approximately one kettle for an afternoon — turned out the register was being read in Wh but the inverter was spitting out values scaled for MWh, giving a 1000x discrepancy. Classic.

Running Venus OS (latest-ish, somewhere in the v3.7x ballpark) and the inverter's own display was perfectly sane — it was purely the SunSpec translation layer having an existential crisis about units.

A few things worth checking if you're seeing this:

• SunSpec model block — confirm which AC energy model your inverter is actually advertising (some manufacturers implement it loosely, shall we say)
• Scale factors — the SunSpec spec has SF registers that are supposed to sort this automatically, but some firmware just... ignores that memo
• VRM vs local display — if the inverter's screen looks right but VRM is showing nonsense, it's almost certainly a scaling issue rather than a metering fault

My setup's a static caravan with a Victron Multiplus-II and a Fogstar battery bank, so accurate yield logging actually matters when I'm trying to work out whether last week's Yorkshire drizzle murdered my solar harvest or not.

Curious whether anyone's cracked a proper fix — custom driver tweak, firmware update, or just accepting the number and multiplying by 1000 in your head like a medieval peasant doing long division?

@DailyWanderer — classic scaling factor issue. SunSpec model 203 has a WH_SF signed scale factor register that some inverters implement incorrectly. If yours is returning SF = -3 when it should be SF = 0 (or vice versa), that's your 1000x discrepancy right there.

Worth pulling the raw Modbus registers directly — I use mbpoll on a Raspberry Pi to read them without the Cerbo GX interpreting anything:

mbpoll -a 1 -r 40093 -c 5 192.168.x.x

Check registers around 40093–40098 depending on your model's SunSpec mapping. The AEG documentation is... not brilliant from what I've seen.

On my narrowboat setup with a Victron Cerbo + third-party inverter I had a similar scaling mismatch — ended up needing a custom driver tweak in Venus OS to apply a manual correction factor until the manufacturer pushed a firmware fix.

@PanelEwan is on the right track but worth being precise — WH_SF applies specifically to TotWhExp (total exported Wh). If you're logging yield rather than export, double-check you're not conflating the two registers. Some inverters report them separately.

Also worth confirming: is the raw Modbus value actually 1000x out, or is the Cerbo's VRM just displaying in MWh when you expect kWh? Had a similar head-scratcher with my Victron setup — turned out to be a unit display toggle buried in VRM settings rather than a genuine scaling fault.

What's your WH_SF register actually returning? If it's 0 when it should be -3, that's your culprit right there.

@PeakExplorer @PanelEwan — both correct, though I'd add that the AEG specifically seems to ship with WH_SF hardcoded to -3 when it should be 0 for some firmware versions. Had exactly this nonsense on my narrowboat setup last winter, Cerbo was reporting I'd generated roughly enough to charge a AA battery.

Worth checking your actual WH_SF register value directly via a Modbus poll rather than trusting whatever the inverter's web UI claims. Tools like mbpoll on a Raspberry Pi will show you the raw signed integer — if you're seeing 0xFFFD that's your culprit (two's complement for -3).

Firmware update from AEG's support page may correct it, though their documentation is, shall we say, aspirationally sparse. Alternatively just apply the correction factor manually in Node-RED before it hits VRM. Bodge? Yes. Does it work? Also yes.

Had this exact saga with my static caravan setup — Cerbo thought my panels had generated enough to power a particularly optimistic LED bulb, turns out WH_SF was sat at -3 when it should've been 0, so divide everything by 1000 and there's your "missing" kilowatt-hours.

Fixed mine by writing the corrected scale factor directly via Modbus TCP using mbpoll on a Raspberry Pi — bit fiddly but doable in an afternoon if you've got the register map handy.

Worth also cross-checking in VRM portal under Advanced → Device List; if the raw register dump looks sane but Venus OS is still mangling it, there's a known quirk where the Cerbo caches stale SF values and needs a full driver restart, not just a reconnect.