Skip to content

Load Forecast (24×7 Profile)

Choose a tag to compare

@hoizi89 hoizi89 released this 22 Apr 09:41

🔮 New: Load Forecast (24×7 Profile)

Your house consumption follows a pattern — mornings, evenings and weekends differ. This integration now learns that pattern automatically and exposes 5 new sensors so you can build smarter battery and charging logic on top of it, alongside your solar forecast and dynamic prices.

✨ 5 new sensors

Sensor Returns
sensor.*_verbrauch_prognose_1h next hour (kWh)
sensor.*_verbrauch_prognose_6h next 6 h
sensor.*_verbrauch_prognose_heute_rest rest of today
sensor.*_verbrauch_prognose_morgen tomorrow full day — with forecast_hourly array (24 values)
sensor.*_verbrauch_prognose_24h rolling 24 h — with confidence_low / confidence_high

All under a new sub-device "[Name] Load Forecast".

🧮 How it works

  • Data source: your existing consumption_entity, pulled hourly from HA's built-in recorder.statistics_during_period API — no external service, no ML, no containers.
  • 24×7 matrix: per (hour-of-day, weekday) cell, rolling mean over 2/4/6 weeks. Monday 07:00 ≠ Saturday 07:00.
  • Modal-drop (on by default): drops the highest + lowest value per cell — cleans out holidays, outages, one-off events.
  • Optional HP/EV subtraction: configure heat-pump or wallbox total-energy sensors → clean base-load forecast.
  • Graceful fallbacks: <14 days → 7-day mean. <3 days → persistence. No data → unavailable.

Typical accuracy (MAPE) at 24 h horizon: 18–30 % — same class as Predbat / EMHASS internals, but zero setup.

🎛️ How to enable

Settings → Devices & Services → PV Energy Management+ → Configure → Lastvorhersage

  1. Toggle Enable load forecast on
  2. Pick History window (4 weeks = sweet spot)
  3. Optionally point to Heat pump / EV charger total-energy sensors

Rebuild runs once per hour in the background. Never blocks the event loop.

🛡️ Safety

  • Fully opt-in — default off. Existing installations see zero changes until you enable it.
  • No breaking changes to existing sensors, config flow, or entity IDs.
  • No new dependencies (requirements stays empty).
  • All errors logged as warnings, never crash the integration.

📊 Dashboard tip (Victron-style chart)

Use ApexCharts with the forecast_hourly attribute of the Morgen sensor + your Solcast tomorrow forecast for the side-by-side bar chart from the feature request that inspired this.

🎯 What's next

Planned for a future release: a built-in binary-sensor that combines this forecast with Solcast + dynamic price for an explicit "charge from grid now" recommendation. v2.0 ships the building blocks — you wire them via templates for now.


Full Changelog: daa9b1b...v2.0.0