Problem
On the live dashboard and day curve, grid power is derived as:
grid = consumption - production + batteryPower
When the consumption meters go unavailable (Zigbee dropouts), their buckets/states report no data and the derived consumption collapses to 0. If the battery is discharging at that moment, the formula yields a large negative grid value, which the UI renders as a big grid export at night with near-zero production - physically impossible.
Real case observed: battery discharging ~3 kW while both consumption meters were unavailable for ~10 minutes. The curve bucket showed consumption = 0 and grid = -3.4 kW after sunset. No actual export happened (grid clamp meter read flat ~70 W import).
Expected
- If any input of the derived grid formula is missing/unavailable for a bucket, the grid value should be suppressed (null) or the point marked as estimated, instead of computing with consumption = 0.
- Same guard applies to computeLive (live view).
Notes
- The per-inverter backup fill (backupPowerId) behaved fine: the gap was filled from the cloud sensor and the curve was marked estimated.
- Battery sign issues in the source sensor can amplify this; separate fix on the HA side.
Problem
On the live dashboard and day curve, grid power is derived as:
When the consumption meters go
unavailable(Zigbee dropouts), their buckets/states report no data and the derived consumption collapses to 0. If the battery is discharging at that moment, the formula yields a large negative grid value, which the UI renders as a big grid export at night with near-zero production - physically impossible.Real case observed: battery discharging ~3 kW while both consumption meters were unavailable for ~10 minutes. The curve bucket showed consumption = 0 and grid = -3.4 kW after sunset. No actual export happened (grid clamp meter read flat ~70 W import).
Expected
Notes