Skip to content

ThermoSmart v1.2.0

Latest

Choose a tag to compare

@github-actions github-actions released this 09 Jul 11:10

ThermoSmart v1.2.0

Stable release following v1.1.0, with a completely rebuilt adaptive Learning Engine, broader device compatibility, safer heating behavior, and improved diagnostics and exports.

Highlights

  • A new adaptive Learning Engine that gradually and conservatively adjusts heating behavior based on observed thermal response, confidence, and outcome quality — always with deterministic TPI control as the safety baseline.
  • Broader, more robust device compatibility across TRV brands and minimal (TRV-only) setups.
  • Additional plausibility checks on outdoor temperature and weather forecast data, so a misbehaving sensor or integration can no longer send implausible values into the heating logic.
  • Safer valve-maintenance behavior that now always respects the Active Control switch.

Learning and control

  • Introduced a new adaptive Learning Engine that observes heating behavior over time and applies small, confidence-gated adjustments on top of the existing TPI controller — never replacing it.
  • Learning starts conservatively and only gains influence as enough reliable data accumulates; low-confidence or early-life zones see little to no adjustment.
  • Learning progress and confidence are now reported in a way that reflects real data maturity rather than optimistic early estimates.
  • Failed or partial heating commands are correctly excluded from what the Learning Engine learns from.

Device compatibility and minimal setup

  • Improved temperature and wind-speed unit normalization across all sensor and weather inputs (Celsius/Fahrenheit).
  • Hardened TRV temperature and command validation for a wider range of device behaviors.
  • Verified and improved support for minimal, TRV-only setups (no external sensors required).

Weather, forecast, and heating safety

  • Outdoor temperature and forecast (high/low) values are now checked for physical plausibility before being used — a garbage or out-of-range reading is treated as missing data instead of distorting the heating decision.
  • Fixed a window-open/close delay recovery issue after a Home Assistant restart.
  • The weekly valve-exercise (anti-seizing) routine now always respects the Active Control switch and no longer runs when Active Control is off.

Storage, recovery, and exports

  • Learning data storage now uses a clearer, neutral internal naming scheme; existing installations are migrated automatically and transparently — no manual action needed.
  • Added richer diagnostic and research export data (storage context, event timelines, daily summaries) to make troubleshooting and voluntary data sharing more useful.
  • Hardened storage retention limits and periodic save behavior for long-running installations.

Configuration and user experience

  • Simplified the zone configuration flow.
  • Clarified sensor and button labeling throughout the UI and documentation.

Tests and quality

  • Added a public automated test suite covering coordinator behavior, weather/forecast handling, device compatibility, and the Learning Engine.
  • Continuous integration now runs Pytest, HACS validation, and Hassfest validation on every change.

Upgrade notes

  • Update via HACS as usual — no manual migration steps are required.
  • Existing zone configurations remain fully compatible.
  • Existing learning data is read and migrated automatically to the new storage format; nothing needs to be deleted or reset.
  • TRV-only setups (without external temperature/humidity sensors) continue to be fully supported.

Known notes

  • The Learning Engine's influence remains intentionally conservative in the early life of a zone — this is expected behavior, not a bug.