Predbat power-flow shows phantom "House" wattage during battery+grid activity — found a workaround, stuck on one setting

[iaing1976]

My setup:
GivEnergy Gen3 hybrid inverter (5kW/9.5kWh battery), GivTCP running in REST mode, Predbat, Home Assistant on Docker. Grid meter connects over a wired RS485-to-Ethernet link. (I did recently swap this from a wireless link to a wired one, but the issue exists on both, so that's not the cause.)

The problem:
When the battery is charging or discharging at the same time as significant grid import/export (like during a force-export event), GivTCP's load_power sensor goes wrong — either near-zero or a huge fake number like 7000W+. My real house load at those times is more like 800-1000W. This then shows up as a nonsense "House" figure on Predbat's power-flow diagram. Seems to be a known weak spot in how GivTCP calculates load in that specific scenario.

What I've tried so far:
Lowered load_filter_threshold from 30 to 5 — this helped Predbat fill in the gaps in its historical load data more accurately (confirmed in the logs), but didn't fix the live display.

Built my own "true load" sensor in Home Assistant using: PV power + Grid power + Battery power. For the grid figure, instead of GivTCP's own (glitchy) reading, I used my myenergi Eddi's independent grid CT clamp — I tested this against a kettle switching on/off and it tracks real power accurately (just a ~10-30 second lag). This sensor stays correct even when GivTCP's own number goes wrong.

Pointed Predbat's load_power setting at my new sensor instead of GivTCP's. Confirmed in the logs this is now being used.

Where I'm stuck:
Even after all that, the power-flow display still shows the wrong number. I've found out why: when GivTCP is in REST mode, Predbat reads power values straight from the GivTCP REST API for the display — it ignores my apps.yaml settings (load_power, grid_power etc) completely, unless givtcp_rest_power_ignore: true is turned on. So my fix helped behind the scenes but didn't change what I actually see on screen.

My questions:
Is turning on givtcp_rest_power_ignore (and setting up all four power sensors — load, PV, grid, battery) the right way to fix the display, or is there something simpler?

Any gotchas to watch for when switching that on in an otherwise REST-based setup?

Has anyone fixed this GivTCP load-calculation issue properly, rather than working around it?

Is using an independent CT clamp (like from an Eddi or Zappi) a sensible way to get a reliable grid reading, or is there a more standard approach people use?

Many thanks in advance for your kind assistance, Iain

Happy to share redacted config snippets if that helps.

[mohammadrezwankhan]

[gcoan]

I would doubt that this is actually a givtcp bug, its almost certainly the case that givtcp reads what the inverter reports as load and passes it through to HA.

The fix is to create your own template load sensor as you have done and to configure rest_power_ignore to force Predbat using the load sensor not REST.

You can use the givtcp grid power sensor as an alternative to your Eddi data and at least will have the data consistently polled from the inverter