IOG dynamic: keep cheap rate to end of 30-min period after car stops charging mid-period #5325

Writen by GLM 5.3 Flash

Setup

• Predbat v9.3.3 (add-on), Octopus Intelligent via the Octopus Energy integration, tariff IOG-SMB (octopus_slot_max: 12)
• car_charging_now wired to a charging-power sensor; car is outside the CT clamp (car_energy_reported_load Off)
• octopus_intelligent_charging On, octopus_intelligent_dynamic On, octopus_intelligent_trust_slots Off

Current behaviour

When the car stops charging inside a dispatch (typically on reaching target SoC), the dynamic check cancels the car's slots once "not charging" has lasted DYNAMIC_LOAD_CAR_SENSOR_MINUTES (2 min, after the 3-min start band DYNAMIC_LOAD_CAR_START_MINUTES). Two things then remove the cheap rate for minutes ≥ now:

• rate_add_io_slots skips its overlay — https://github.com/springfall2008/batpred/blob/v9.3.3/octopus.py#L3730
• dynamic_load_car_strip_feed_rates overwrites the discount already delivered by the integration's feed (is_intelligent_adjusted), setting those minutes back to peak — https://github.com/springfall2008/batpred/blob/v9.3.3/octopus.py#L3792 — but only minutes ≥ now, outside the fixed 23:30–05:30 window, and not covered by another car's trusted dispatch

Elapsed minutes keep their cheap rate.

This isn't specific to confirm-first mode: the cancellation in dynamic_load_car_target() (https://github.com/springfall2008/batpred/blob/v9.3.3/plan.py#L487) fires for any car inside a dispatch that stops charging, regardless of octopus_intelligent_trust_slots — that switch only controls the pre-confirmation default. I run confirm-first, but the fix would apply to the dynamic check generally.

Why this looks wrong

Predbat's own rate model already treats any partial overlap with a 30-minute period as the whole period off-peak: rate_add_io_slots floors the start and ceilings the end ("to ensure any partial overlap with a 30-min slot marks the entire slot as off-peak", https://github.com/springfall2008/batpred/blob/v9.3.3/octopus.py#L3712), and dynamic_load_car_strip_feed_rates likewise rounds out to whole periods (https://github.com/springfall2008/batpred/blob/v9.3.3/octopus.py#L3812). My working assumption is that Octopus prices the same way per half-hour settlement period — I can attach a bill/consumption export if needed. On that assumption, when the car stops part-way through a period, the rest of that period is cheap in reality but modelled at peak, and Predbat won't plan the house battery into minutes that are actually cheap. The cancellation is finer-grained than the rate model's own period granularity.

Example

Dispatch 20:00–21:00, car reaches target and stops at 20:12:

Request

Once charging has been seen within the current 30-min period, keep the house cheap rate until that period ends, i.e. snap the strip/cancel effective time to the next period boundary after charging evidence.

• Periods with no charging evidence keep today's behaviour, so the GH#5080 protection against unused dispatches is preserved
• The car's modelled charging and "Hold for car" should still stop when the car stops; only the house rate persists
• The fixed 23:30–05:30 window is unaffected
• Happy for this to sit behind a switch if a default change is too opinionated

Scope

At most one 30-minute period per stop event: only the tail of a period where charging stopped mid-period.