Precision loss in total export energy (SunSpec Model 101)

[kereis]

mbmd uses the SunSpec Model 101 for reading total export energy in Wh.

For SunSpec registers, Model 101 and Model 122 can be used. However, it looks like that the total export value in Model 101 is slightly unprecise.

Here you can find an example. You can simply check it out (https://github.com/LibreWB/mbmd/commit/ca23429878711eb0195bf98e254017a7eaa00c35) and run mbmd scan -c <CONFIG_FILE> -a <YOUR_ADAPTER>.

https://github.com/LibreWB/mbmd/blob/ca23429878711eb0195bf98e254017a7eaa00c35/meters/sunspec/sunspec.go#L191

actWh, actWhErr := client.ReadHoldingRegisters(40302, 4)
wh, whErr := client.ReadHoldingRegisters(40209, 2)

if actWhErr == nil {
    fmt.Printf("ActWh > %d\n", binary.BigEndian.Uint64(actWh))
} else {
    fmt.Println(fmt.Errorf("actWhErr > %s", whErr))
}

if whErr == nil {
    fmt.Printf("WH > %d\n", binary.BigEndian.Uint32(wh))
} else {
    fmt.Println(fmt.Errorf("whErr > %s", whErr))
}

What we do here is reading from the SunSpec register directly via ModBus client. Total Export Wh in Model 101 is 32-bit, in Model 122 64-bit. Actually, this shouldn't make a big difference as both total export values should fit into each sizes. However, it looks like that the 32-bit value is rounded up to the next decimal place. Example values are:

ActWh > 23665022
WH > 2366502

It might not be a huge problem for people who use mbmd casually, but in terms of monitoring, one decimal place could make a difference - at least for our current implementation of Prometheus (#195).

I've already chatted with @andig about this "issue". We should maybe look into this and think about maybe adding a model mapping entry for Model 122 or replace it with 101.

I don't know if this affects any other values, but we noticed the value difference in total export energy.

[andig]

Hat WH einen Scale Factor? Evtl. dividieren wir hier einfach blöd? Oder kommt es wirklich falsch vom Bus?

[kereis]

Das sind Werte, die direkt vom Register rauskommen - ich habe da nichts umgerechnet und auch kein Scaling Factor angewandt. Im SMA-Register kommt das gleiche wie bei ActWh raus.

EDIT: Selbst wenn wir teilen würden, ändert es nichts an der Tatsache, dass da eine Stelle bei WH fehlt.