To understand what happened there, you need to inspect the actual schedules produced by the DESS Trade algorithm in the hours leading upto the moment the new day ahead prices roll in (around 14h in the graph). Until that moment of the day, DESS will just target MinimumSoC at midnight. Unfortunately those schedules cannot be inspected afterwards because they get overwritten by the actual performance data, you’d have to screenshot them as they happen in real time.
Apparently the algo saw enough ‘free’ solar energy later this day to charge up the battery to full, and subsequently discharge to minimumSoC, before midnight.
And subsequently discarded to evaluate the ‘edge case’ possibility of ‘cheaper then free’ grid energy. Until the new prices rolled in around 14h but then it was too late.
There is a recurring theme I coined ‘downward selling bias’ originating from a standing assumption that the optimal ‘Scheduled SoC to target at the end of the known day-ahead pricing period’ always coincides with the battery’s technical/safety ‘MinimumSoC’ parameter.
The issue with that assumption is that it is based on historical observations of systems where installed inverter and solar power in relation to battery capacity was predominantly large and nett negative price conditions were rare enough to ignore. Under those conditions the optimal ‘end of known price period’ SoC to target, indeed trends towards MinimumSoC.
But with larger batteries and less or no solar power, these conditions have become increasingly less common, which is clear from the increase of topics ‘why doesn’t DESS Trade buy cheap from grid during the early hours of the day (upto the moment next day prices roll in)’. And I expect that to further increase when ‘salderen’ stops end of this year because that will lead to an further uptake of large battery capacity installation.
The correct solution is to decouple that ‘Scheduled SoC to target at the end of the known day-ahead pricing period’ from ‘MinimumSoC’ altogether. Even though it seems unlikely Victron is going to do so, one can hope they’ll ‘see the light’ nevertheless.
Me and quite a few others are therefore now using a workaround to achieve practically the same: we set a large, dayly forward rolling ‘virtual’ EV consumption forecast adjustments at midnight. Virtual because it is never actualized, it gets rolled to the next day before it ever gets reached. The effect is that the scheduler targets an adjustable higher SoC at 23h. And that enables the scheduler to start charging early enough in the morning not to miss the low price buy opportunities.
All the above is unrelated to the announced ‘turn off solar when prices are nett negative’, I’m afraid that’s not going to solve your observed behavior.
