Is it not charging in vain?

[ridax67]

The spot price now is 102 and will top at 111 during evening. Is it correct to continue charging? Wouldn’t it be better to give it up and start exporting, taking the 40 battery cost in consideration? Or am I thinking wrong?

[ridax67]

bess-debug-2026-07-17-183806.md

[johanzander]

At 18:30 the battery wasn't charging — the schedule shows SOLAR_EXPORT with zero battery action, and the battery was already ~87% full (17.4/20 kWh), charged almost entirely from solar earlier in the day (essentially no grid import went into it). The plan for the rest of the day already had zero further charging scheduled.

What the optimizer actually did was hold that stored (largely free) energy and discharge it starting at 20:00, right through the evening peak — including the exact 111 öre hour you flagged at 21:00, where it exported 2.5 kWh for about 3.3 SEK of savings in that single 15-minute slot. So the plan you're asking about ("give up charging, export at the good price instead") is exactly what was already scheduled — it just hadn't kicked in yet at 18:30 because the peak was still 1.5 hours away.

[ridax67]

But I have the battery cost set to 40. Charging in the afternoon and discharge a few hours later makes a charge discharge cycle. The top during the evening was 111 and it means it shouldn’t charge when the price is above 71, but it should export it immediately rather than charge the battery and sell at a higher cost but incurring battery cycle cost.

I don’t have yesterdays pricing at hand now, but I am sure the price was well above that long before 18 a clock.

[ridax67]

I am not talking about battery discharge, that it did at the correct time. But when pricing isn’t 40 below top, it should switch to solar export instead, shouldn’t it?

[johanzander]

Good follow-up — let's separate the two decisions, because the DP doesn't use "top price minus 40 öre" as its rule, and I want to show you the actual numbers so it's clear why.

The cost of charging from solar instead of exporting it isn't compared against the day's peak buy price — it's compared against the sell price you'd have earned by exporting that solar right then, plus the 0.40 SEK/kWh cycle cost.

On the actual profitability: the battery stored ~13.3 kWh of solar that afternoon at an average opportunity cost of ~1.22 SEK/kWh (forgone sell price plus the 0.40 SEK/kWh wear cost, both already netted into the system's own per-period savings figures). Tracing the actual discharge through the entire overnight drawdown (not just the 20:00–23:45 window) shows ~14.6 kWh discharged by 06:45 the next morning — 6.8 kWh avoiding grid imports at ~1.84 SEK/kWh and 7.8 kWh exported at ~1.21 SEK/kWh, for a blended ~1.50 SEK/kWh. Net across the full charge/discharge cycle: about +5.7 SEK profit, ~0.28 SEK/kWh margin over cost. (Note: the post-midnight portion was the optimizer's committed plan as of 18:30, not yet-realized actuals at export time.)

The other piece: the DP isn't checking a single static threshold ("below top minus 40") at all. It's a full backward pass over the whole day's price path — at every 15-minute period it weighs "forgo selling now vs. hold and discharge at some future period" against everything that comes after, not against one peak number. That's why you'll see it store at several different sell prices through the afternoon rather than picking one cutoff.

[ridax67]

But I am not looking at what mathematical model that can explain the behaviour. I understand there are reasons behind everything, but are they the correct reasons? Could another scheduling result in a better result? In this case it is my opinion that, yes it would. When the next day prices were published at 13 a clock it was clear that there would be no point in storing solar for using the next day. So then there is only what is left of today that are of interest. The prices were not low during the afternoon, as far as I recall. Disregard the algorithm and look at the case and it should be clear that the afternoon charging shouldn't have been done, but it should have been exported directly.

Your model is very good at finding the most economical way to charge, but it seems to lack the ability to see when it shouldn't do it, or should not complete it when tomorrows pricing becomes clear.