EcoEdge

Weather-aware temperature predictions — measured honestly

EcoEdge predicts how each of your rooms will behave over the next 24 hours, hour by hour, driven by the weather forecast and your room's measured thermal model. And — unusually for this industry — we score those predictions against reality every week and publish the honest result.

How the prediction works

Every room's forecast combines three ingredients:

  1. Your room's physics — the measured heat-loss coefficient (k) and heating response, fitted from your own thermostat history. See building heat-loss calculation.
  2. The weather forecast — hourly temperature, wind, cloud cover, and solar radiation for your home's exact coordinates.
  3. Cross-home machine learning — models trained across the fleet that refine the physics when weather patterns carry extra signal (a windy cold snap behaves differently from a calm one at the same temperature).

The result appears on your dashboard as the dotted forecast line: predicted indoor temperature and the recommended setpoint for each coming hour.

The scoreboard: predictions vs. reality

Anyone can draw a confident-looking forecast line. The question that matters is: was it right? EcoEdge answers with weekly backtests — the same discipline used in quantitative finance:

  • Rewind the clock and let the model predict, using only data available at that moment.
  • Compare against what the thermometer actually recorded 30, 60, and 180 minutes later.
  • Score it against the honest baseline: persistence, the assumption that the temperature just stays where it is. Any model that can't beat persistence has learned nothing.

The results — mean absolute error per horizon, per home — are what drive development decisions. When a model loses to the baseline, that's recorded, and the model doesn't ship. When it wins, the margin is measured, not asserted.

Why predictions matter for your heating bill

Prediction is what turns a thermostat from reactive to anticipatory:

  • Pre-heat with purpose — if the forecast shows the afternoon sun will warm the living room by 1.5°C, the heating doesn't need to do that work.
  • Coast into setbacks — a well-insulated room can stop heating early and glide to the night setback on stored heat.
  • Know your recovery time — the model knows a bathroom with k = 0.7 needs 40 minutes to recover two degrees, so morning warm-up starts exactly early enough and not an hour too soon.

Every recommendation stays advisory: EcoEdge never changes your thermostats. You see what the model would do and why — the gap between your setpoint and the recommendation is where the savings live.

Create a free account to see 24-hour predictions for your own rooms, or read how to connect Home Assistant.