v8.41.0

What's Changed

New: automatic weighted load forecast (days_previous_auto)

• New load forecast code by @springfall2008 in #4101

Predbat can now predict your house load using a weighted average of all your recent history instead of a fixed set of days_previous days. This is more robust when you have gaps in your history or your usage changes (e.g. returning from holiday), as it no longer relies on a few specific days being present and representative.

Enable it in apps.yaml:

  days_previous_auto: True

What it does

For each 5-minute slot in the plan it looks at the same time of day across your history and combines those samples into a weighted average. Each historical sample is weighted by how relevant it is:

• Weekday – same weekday as the slot counts fullest (1.0); a different weekday but same type, i.e. weekday-vs-weekday or weekend-vs-weekend (0.7); weekday-vs-weekend (0.5).
• Holiday – samples recorded in a different holiday-mode state to today are halved, matched per 5-minute sample (so a mid-day change is handled).
• Age – yesterday counts most (0.9), reducing by 0.03 per day down to a floor of 0.1, so recent days dominate.

Slots with no recorded data (gaps) are ignored so they don't drag the estimate down.

How much history is used

The search window is taken from max(days_previous) (or 7 days if days_previous isn't set), capped at 30 days. So to forecast from the last two weeks of history:

  days_previous: 14
  days_previous_auto: True

If Load ML is enabled it takes precedence over days_previous_auto.

Full Changelog: v8.40.16...v8.41.0