programmatically identifying when the battery heater is active

paulshields · Fri Nov 13, 2026 10:39 am

I'm looking at capturing when the heating element comes on for my EP12s, just so I can chart it in HA and grafana.

From what I can gather via scouring the source code the Fox ESS Modbus integration doesn't expose the heater. For now I've written a sensor in HA which basically does this "if battery low temp rises quickly and plant room temp does not then heater on = true". There's a bit more to it than that (thanks Claude) and probably not 100% accurate but it worked for the first time with last night's cold snap.

I don't want to call the Fox Cloud API (assuming it's exposed via that).

Anyone else have their own mechanism or a better suggestion? I was going to ask on the Fox integration repo but thought I'd check here first.

MattyS · Fri Nov 13, 2026 3:42 pm

For reasons unknown, although the heater is exposed by modbus for reading purposes the only way to write the heating changes afaik is via the API. There is a HA integration called FoxESS Plant that I've found today and that uses the API to make the changes, but without leaving HA.

paulshields · Fri Nov 13, 2026 5:01 pm

Oh so you can get status of whether the heating element is exposed via Modbus via the Fox ESS integration? That was all I was after but I don't see it. I don't need write access at all. My bodge using inference does seem to be working ok but I'd rather have empirical proof.

nrb501 · Fri Nov 13, 2026 7:10 pm

Another method possible when using the "nathanmarlor/foxess_modbus" integration is to compare the entities "Battery Current" and "Inverter Battery Current". One will be a positive value and the other a negative, normally summed together they are about zero, when heating is active the sum will show about a 1 Amp difference per EP11. You can see on the history graph from my system (2 EP11s) the difference both during charge and discharge when the heater is active.

paulshields · Fri Nov 13, 2026 8:35 pm

Ah that's a brilliant way of capturing this. I've chatted to my LLM about your method and it agrees, and Claude is (hardly) ever wrong

. Thanks for that suggestion. I'm going to run it in parallel with my other method for a bit, but from looking at historical data it seems to map precisely.