Observer sensor / inner control loop

[baal86]

Had a discussion with a friend very knowledgeable about controls this weekend over a few beers. Since our overall problem seems to be slow power meters (HA, TASMOTA, HTTP) and erratic behaviour of batteries, maybe a second, very fast auxiliary power sensor (shelly 1PM) at the output of the inverter could be used to simplify the control loop. This could work like that:

• Inner loop uses fake CT002 power meter and all the tricks for balancing, efficiency, saturation... to bring the auxiliary power meter to a defined set-point, not 0.
• Outer loop slowly reads the actual grid import-export power and adjusts the set point of the inner loop at a speed that it can follow.

I am not sure how effective this will be in practice. It might help people who cannot install a physical power meter at the grid connection point (i.e. likely most AstraMeter users). Since AstraMeter is quite a complex thing right now, maybe this is something to consider for a future major refactoring since it will interact with everything we have right now.

[tomquist]

What you're suggesting is essentially already how it works. AstraMeter doesn't run the control loop itself. It feeds grid power to the Marstek device, which does the fast inner regulation. The "outer loop adjusting a setpoint" behavior is what the filter chain is for, and it's all configurable today:

• PID: PID_KP / PID_KI / PID_KD (start at PID_KP = 0.5), plus PID_MODE and PID_OUTPUT_MAX
• Setpoint / import buffer & scaling: a small negative POWER_OFFSET, and POWER_MULTIPLIER
• Outlier rejection (Hampel): HAMPEL_WINDOW, HAMPEL_N_SIGMA, HAMPEL_MIN_THRESHOLD
• Smoothing: SMOOTH_STEP, SMOOTH_TARGET_ALPHA
• Update rate: THROTTLE_INTERVAL

Worth tuning those against a slow meter before considering any refactor.

[baal86]

Yes! There are many options and it is stable for me using the "bias" PID with basically any values for "P" and I=D=0.

What I was thinking was to provide AstraMeter a second sensor at the inverter to measure actual power generated by batteries + solar. This could improve response time and also shield the control loop from aggressive load steps on the grid connection better than the current filters can since they will always act by effectively making feedback slower.

Since this is incompatible with the current implementation of AstraMeter I would not classify that as a feature request but rather an idea to keep in mind if you ever decide to refactor it anyways.

[tomquist]

We already get this. In the CT002/CT003 protocol, every battery reports its own output power in its request. The active-control balancer uses it, that's what drives fair distribution and saturation detection, and that's why the CT002 emulator is generally better than what you can achieve with a Shelly emulator.

[baal86]

Ah! OK, if that reporting is accurate then this is indeed pointless for the CT00x. Does not take into account inverter effects but I guess comparing to the batteries those don't matter