SPH 3-6kW / 7-10kW / 8000-10000TL-HU: register 1021 may be grid import, not load power (needs field confirmation)

[0xAHA]

Summary

Registers 1021 and 1037 look mislabelled on the legacy SPH profiles. If that's right, "Load Power" on those profiles is actually showing grid import, and the real load power is read and thrown away.

I have not changed anything. I need one confirmation from a real device first, and it costs about a minute.

✅ Confirmed — no further checks needed

@OmkoH confirmed this on an SPH 5000 (chart): Power to Load, Power to User and Grid Import Power all read 747 W simultaneously, tracking exactly across a sunny August morning down to a shared 29 W baseline. Genuine house load would have diverged from grid import the moment solar or the battery carried the house.

Register 1021 is grid import. The mapping is corrected on all three profiles:

Register	Was	Now
1021/1022	power_to_load	power_to_user_total — grid import
1037/1038	self_consumption_power (unused, discarded)	power_to_load — house load

If you're on one of the affected profiles, Load Power will begin reporting your actual house consumption rather than grid import, so it will stop dropping to zero while you run on solar or battery. Self Consumption is derived from Load Power and becomes correct as a result. Existing history is untouched; only new readings change.

Thanks to @ian-mcarthur-oxford, @alanmk and @De4dCert as well — no need to check anything now, this is settled.

The original check, kept for the record

Chart these two entities, both on the Load device:

Entity	Source
Growatt Load Power to Load	register 1021 — the one in question
Growatt Load Power to User	register 1015

Identical lines mean the mapping is wrong. Both are read directly from the inverter, so neither is derived from the other.

Grid Import Power is not a valid third comparison — it isn't read from the inverter, it's calculated from Power to User, so it matches that entity regardless of what 1021 contains. The original wording of this issue offered it as an alternative, which was a mistake on my part.

Which profiles

Profile	Affected
SPH 3-6kW	suspected
SPH 7-10kW	suspected
SPH/SPM 8000-10000TL-HU	suspected
SPH ... (V2.01) variants	not affected — already correct
SPH-TL3, SPA	not affected — already correct

Check under Settings → Devices & Services → Growatt Modbus → Configure which one you're on.

Why I think this

The storage range follows a regular pattern of four 8-register blocks: R phase, S phase, T phase, then Total. Our own docs/developer/protocol-v139.md, extracted from the Growatt Modbus Protocol II documentation, gives:

1015  Pactouser R H       AC power to user R-phase
1021  PactouserTotal H    AC power to user total        <- grid import
1029  Pactogrid total H   AC power to grid total        <- grid export
1037  PLocalLoad total H  INV power to local load total <- house load

The three legacy SPH profiles instead map 1021 as power_to_load and 1037 as self_consumption_power — and self_consumption_power isn't in any sensor group, so 1037 is read and discarded. Four independent sources say 1021 is grid import:

1. The protocol documentation above.
2. The SPA profile, field-verified in SPA 3000TL BL not mapping correctly #249 by energy balance (battery discharge + grid import = load held across every scan).
3. Our own SPH-TL3 and SPH V2.01 profiles, which already map it correctly.
4. The Homey Growatt app, an independent implementation: pactousertotal: [1021, 'import from grid to user total'], totalhouseload: [1037, 'Total house Load'].

Why the symptom is subtle

On a single-phase inverter, R phase and Total are the same measurement. So register 1015 (power_to_user, mapped correctly) and register 1021 (power_to_load, suspected wrong) would return identical values — which is exactly the signature above.

Both numbers look plausible on a dashboard. Grid import is a real quantity; it just isn't your house load. The likely visible symptom is Load Power sitting at zero whenever you're not importing — during a sunny afternoon, or overnight on battery — while the house is obviously still consuming.

If you've ever wondered why Load Power drops to nothing while your battery discharges, this would explain it.

What happens next

If it's confirmed, the fix is straightforward: 1021 becomes power_to_user, 1037 becomes power_to_load, matching what the V2.01 profiles already do. It will change what your Load Power sensor reports, so I want evidence before touching profiles this widely used rather than after.

If it's not confirmed on your firmware, that's equally valuable and I'll say so here.

---

Tagging a few people who've reported on single-phase SPH systems, in case you're willing to look — no obligation, and the check is genuinely a one-minute glance at History:

@OmkoH — your #326 is what added these registers to the 3-6kW and 7-10kW profiles, so you're the closest thing to a primary source here.
@ian-mcarthur-oxford — you looked at SPH grid power flow in #302, so you'll know immediately whether these two track each other.
@alanmk — SPH 3-6kW on DTC 3502 from #337.
@De4dCert — SPH 3600TL BL-UP from #333.

If you're on an SPH-TL3, SPA, or a V2.01 profile, this doesn't affect you.

[OmkoH]

Hi 0xAHA, the following text is with the help of Claude. I hope this helps.

@0xAHA I have an SPH 5000 single-phase setup. I checked the Home Assistant
History chart with the requested entities:

• Growatt Load Power to User
• Growatt Grid Grid Import Power

Result: The two lines are identical – they track each other perfectly
throughout the graph, including during different times of day.

This confirms your suspicion that register 1021 is grid import, not load power.

[0xAHA]

Thank you for taking the time — but I have to stop you before this goes further, because my instructions were wrong and that chart cannot answer the question. The fault is entirely mine.

Why those two entities are always identical

You charted Power to User and Grid Import Power. Grid Import Power is not read from the inverter at all — it is calculated, and it is calculated from Power to User:

elif self._sensor_key == "grid_import_power":
    import_power = getattr(data, "power_to_user", 0)

So the two lines are the same number by construction. They would track each other perfectly on every inverter ever made, whether or not register 1021 is mislabelled. Your chart is correct; it just measures nothing.

Worse, the entity actually under suspicion — the one fed by register 1021 — is not on that chart at all.

My original post asked for "Load Power", which is not the name of any entity, and then offered "or Grid Import Power" as an alternative, which is the one choice that could not work. That is a bad instruction and you followed it reasonably.

The check that does work — and it needs only one entity

Forget pairs. Look at a single entity:

Growatt Load Power to Load

Then answer one question:

Does it sit at zero while your house is obviously still using electricity?

The moments to look at are a sunny afternoon when you are exporting, or overnight when the battery is discharging and you are importing nothing. Your house is drawing power at both times.

• Reads zero when you are not importing from the grid → register 1021 is grid import, the mapping is wrong, and I will fix it
• Shows a sensible household figure regardless of grid direction → the mapping is correct on your firmware, which is just as useful to know

A screenshot of that one entity over 24 hours settles it.

If you would rather do the two-entity version

Both live on the Load device, and their names are unhelpfully similar — this is the trap I walked you into:

Chart this	Not this
Growatt Load Power to Load	Growatt Grid Grid Import Power
Growatt Load Power to User

If those two are identical, the mapping is wrong. Both are read straight from the inverter, so neither is derived from the other and the comparison means something.

A note on the method

I asked for a comparison between two values where one is computed from the other, then treated matching lines as confirmation. Two readings agreeing proves nothing on its own — it is the diverging case that carries information — and I should have checked what Grid Import Power was before naming it in the test.

I would rather find this now than after changing three widely-used profiles on the strength of it. Your Load Power sensor is unchanged and nothing has been altered.

Sorry for the wasted effort, and thank you for coming back with something specific enough that the mistake was visible.

[0xAHA]

Ignore my previous comment — it was wrong, and your test was right. I read your bullet list and never opened your screenshot. Your chart has three entities on it, not two:

Growatt Load Power to Load:      747 W
Growatt Load Power to User:      747 W
Growatt Grid Grid Import Power:  747 W

Power to Load is there. That is the entity fed by register 1021, it is the one I needed, and it tracks Power to User exactly — including the 29 W baseline across a sunny August morning, when a genuine house load should have diverged from grid import the moment the sun or the battery carried the house.

That is the confirmation. I talked myself out of it by skim-reading the reply that contained it, which is worse than the original ambiguous instruction.

Confirmed, and fixed

Register 1021 is grid import, not load power. Corrected on all three affected profiles:

Register	Was	Now
1021/1022	power_to_load	power_to_user_total — grid import
1037/1038	self_consumption_power (unused, discarded)	power_to_load — house load

This matches the SPH V2.01, SPH-TL3 and SPA profiles, which already had it right, and the V1.39 layout of four 8-register blocks (R, S, T, Total).

What changes for you when this ships: Load Power will start showing your actual house consumption instead of grid import — so it will no longer drop to zero when you are running on solar or battery. Self Consumption is calculated from Load Power, so that becomes correct too, having been derived from grid import until now.

Your existing history for those sensors stays as it is; only new readings change.

What your chart proved that I could not have found otherwise

Four documents said 1021 was grid import. None of them could tell me what your firmware actually does, and I was not willing to change three widely-used profiles on documentation alone. One chart with the right three entities settled it in a way nothing else would have.

Thank you for going and getting it — and my apologies for the round trip. The instruction I originally wrote was ambiguous, you interpreted it correctly anyway, and then I second-guessed you without looking properly.

[OmkoH]

Hi
Just to be complete I have included screenshots of august 6. the battery is discharching and about 3000 watt is going out of my house.
thanks for the great work you are doing!

[0xAHA]

@OmkoH thank you — and the second screenshot is the one that settled it, so I am glad you sent it.

What your screenshots proved

The Home Assistant chart shows three entities that should be three different quantities reading exactly the same number:

11 August 2026, 10:34:07
  Growatt Load Power to Load:     747 W
  Growatt Load Power to User:     747 W
  Growatt Grid Grid Import Power: 747 W

On its own that is suggestive but not conclusive — entities agreeing can be coincidence.

The ShinePhone screenshot is what made it conclusive. At that moment your inverter's own portal reported:

Load power	2.40 kW
Import	0.00 kW
Discharging	0.60 kW

Load and import are plainly different quantities in Growatt's own app, while Home Assistant was publishing them as one number. Two values diverging proves they are separate sources in a way that two values agreeing never can.

So register 1021 was grid import, exactly as suspected, and the real house load was somewhere else.

The fix

Corrected against the protocol document, which turned out to state it plainly:

Register	Documented as	Was mapped as	Now mapped as
1021/1022	PactouserTotal — AC power to user total	Load Power	Grid import
1037/1038	PLocalLoad total — INV power to local load	Self-consumption	House load

Applies to the SPH 3-6kW, 7-10kW and 8000-10000TL-HU legacy profiles.

Please install v1.6.1 to confirm — you will not be offered it automatically

Your screenshots are from before the fix existed, so they cannot tell us whether it worked. The fix is in v1.6.1, which is a pre-release — HACS will not show it unless you opt in:

1. HACS → Growatt Modbus → three-dot menu → Redownload
2. Tick Show beta versions
3. Choose v1.6.1
4. Restart Home Assistant

What to look for

The same chart you already sent, over an hour or two of normal use. Before, the three lines sat exactly on top of each other. After, they should separate:

• Load Power should follow the "Load power" figure in ShinePhone (2.40 kW in your screenshot)
• Grid Import Power should follow "Import" (0.00 kW in your screenshot)

If they still track each other identically, the fix has not worked on your model and I would want to know that just as much.

A screenshot of that chart is all that is needed — no scan or diagnostics.

One caveat worth stating

v1.6.1 is a pre-release because a separate change in it, affecting MOD/MID three-phase models, has not been confirmed on hardware yet. Nothing in it touches SPH beyond this fix, and v1.5.5 remains the stable release if you would rather wait — this fix will reach stable once both pieces are confirmed.

Thanks again for the follow-up screenshots. Sending the portal view alongside the Home Assistant one is what turned this from a plausible theory into something I could act on.

[0xAHA]

@OmkoH good news, and it removes the awkward part of my last message: v1.6.2 is now the stable release, so the 1021/1037 fix reaches you through a normal HACS update. Ignore the "Show beta versions" instructions I gave you — you no longer need them.

The check is unchanged and still just one screenshot: the same history chart, over an hour or two of normal use. Before, your three entities sat exactly on top of each other at 747 W. After, Load Power and Grid Import Power should separate, with Load Power following ShinePhone's "Load power" figure and Grid Import following "Import".

If they still track each other identically, the fix has not worked on your model and that is just as important to know.