Perfidus

Update: 
Output offset is the result of bad choice for opamp, with large figures for input offset voltage. Because of the gain setting in the OA, 10kohm/100ohm =100x, 10mV of input offset = 1 V output shift.
Tested with some improvised compensation circuit, and managed to pull the output of U7 to zero voltage, for the DSP to think that it has zero PV current. The fans stopped, and after 15 minutes the PWM circuit deactivated (no more back feed of battery voltage into solar panels) and "PV charging" with current drain for battery clearly stopped, putting the inverter to OFF, so the firmware does its job properly.
Now I order a new op-amp,  I selected OPA228 from TI, with 10uV input offset voltage and optional external offset adjust inputs.

I note that Axperts (all models, as far as I know) do not have a direct battery current measurement. Instead they calculate/infer/guess it by other measurements, and making various assumptions/guesses for losses. So it's not too surprising to see that the accuracy goes up with larger loads.
The  same place you find all the best information for Axperts 😎: Pip inverter repairs and hardware modifications. There are contributions from many people from all over the world, including a few from yours truly. The above link goes straight to the schematics  index, but there are sections for repair, components, etc etc.