Skip to content

predheat_calibration.md

Latest commit

 

History

History
148 lines (128 loc) · 10.1 KB

predheat_calibration.md

File metadata and controls

148 lines (128 loc) · 10.1 KB
Raw

Predheat calibration (offline, winter 2025/26)

Correction (8 Oct 2026) — supersedes heat_loss_degrees 0.0315 and heat_cop 3.81 below. Free-cooling nights show the building time constant is ≈ 60 h, not 32 h (C ≈ 5.5 kWh/K, see docs/capacity_learning.md); the winter daily balance had under-estimated C. Replay over 150 winter days (tools/predheat_replay.py functions, all days):

Set Elec MAE Elec bias Room MAE @6 h @24 h
current (156, 0.0194, gains 400, COP 3.68) 0.85 +0.24 0.66 2.53
earlier advice (94, 0.0315, 440, 3.81) 0.91 −0.27 0.78 1.10
94, 0.0171 (C 5.5), COP 3.81 1.12 −0.82 0.36 1.11
recommended: 94, 0.0171, 440, COP 3.2 0.82 +0.04 0.36 1.11

With the correct C the room forecast halves its 6 h error, but Predheat then under-predicts electricity (its structure lacks RT modulation and the circulation pump/standby inside the COP); heat_cop 3.2 is an empirical calibration of the energy level (lower gains or higher heat_min_power worsened the room forecast without removing the bias). Recommended predheat: values:

    heat_loss_watts: 94
    heat_loss_degrees: 0.0171
    heat_gain_static: 440
    heat_min_power: 680
    heat_cop: 3.2

(keep the heat_pump_efficiency table, flow_difference_target 5 and hysteresis 1.0 / 1.0).

Update (M1): the values below were validated by replaying Predheat over 150 winter days (tools/predheat_replay.py, report in docs/predheat_replay_report.md). The recommended values are the "fitted" set in that report, which supersede the first-pass table below where they differ (heat_loss_degrees 0.0315, heat_min_power 680, heat_cop 3.81 with a full heat_pump_efficiency table). Hold-out result versus current settings: daily heating-electricity error 0.98 → 0.84 kWh/day, bias +0.40 → −0.11 kWh/day, hourly room-temperature RMSE 1.90 → 1.16 K, 24 h room error 2.56 → 1.16 K.

A further "calibrated" set fits room temperature better (RMSE 0.75 K) but pushes several parameters to their search limits and over-predicts electricity by ~0.5 kWh/day; since Predbat consumes the energy forecast it is not recommended. The remaining room-temperature error is structural: Predheat does not model the Daikin's RT modulation or the overnight deviation schedule.

hysteresis/hysteresis_off stay at 1.0. The heating_energy fix below still applies.

volume_temp note (verified in Predheat/Predbat source): Predheat reads volume_temp with get_arg("volume_temp", …) without domain="predheat", and Predbat's get_arg then looks only at the top level of pred_bat: (or HA config), so a volume_temp: inside the predheat: block is ignored. Per the Predheat docs the sensor must measure radiator water near the room thermostat, not at the heat pump, so the Daikin return sensor (R4T) is not a suitable substitute (earlier suggestion withdrawn). Use the living-room Zigbee radiator sensor, placed at the top level of pred_bat:, once it is back online; until then Predheat falls back to its simulated input_number.predbat_next_volume_temp.

First pass (superseded where noted)

Based on the daily/hourly fits in docs/data_inventory.md and the current predheat: section of apps.yaml (shared 7 October 2026). Predheat semantics verified against apps/predbat/predheat.py (batpred main):

  • heat_loss_watts = UA (W/K); watt_per_degree = heat_loss_watts / heat_loss_degrees is the building heat capacity C (Wh/K).
  • heat_energy / $external is cumulative electrical energy (heat_power_out / (heat_cop · cop_adjust)), keyed every 10 min.
  • COP: heat_cop × heat_pump_efficiency[outdoor °C] / max(table).
  • Heat output modulates between heat_min_power and heat_max_power in proportion to (flow_temp − volume_temp) / flow_difference_target.

Suggestions only. Nothing has been changed in HA; apply manually and compare forecast error over the following weeks.

Parameter comparison

Setting Current Fitted / observed Suggested Confidence Effect of current value
heat_loss_watts 156 UA 94 ± 3 W/K (daily balance, R² 0.90) 95 high Heat demand over-predicted by ~65 % at design conditions
heat_gain_static 400 440 ± 34 W 420 medium fine
heat_loss_degrees 0.0194 → C = 8.0 kWh/K C ≈ 3.0 kWh/K (daily 3.0 ± 0.8; hourly 2.7–3.2) 0.030 (→ C ≈ 3.2 kWh/K) medium Room responds ~2.5× too slowly in the forecast
heat_output 10630 K·50^1.3 ≈ 10.2 kW (K ≈ 63, n ≈ 1.3–1.38) keep 10630 high consistent with radiator data
heat_volume 121 not identifiable from data keep – –
heat_max_power 4000 EDLA04E2V3, 4 kW class keep medium –
heat_min_power 500 min. modulated output ≈ 680 W (P5), median 900 W 750 medium Under-predicts cycling in mild weather
heat_cop + efficiency table 3.68 with default table (≈ 2.5 at 0 °C, 2.95 at 5 °C, 3.35 at 10 °C) measured daily COP 2.8 (≤ 3 °C), 3.0 (3–6), 3.3 (6–12), 4.0 (12–16) heat_cop: 4.0 + table below medium (1 kWh counter resolution) Slightly pessimistic in cold weather
flow_difference_target 5 observed flow−return ΔT ≈ 1.9 K 3 low Predheat-internal tuning parameter; check forecast shape
hysteresis / hysteresis_off 1.0 / 1.0 Daikin room hysteresis 9-0C = 1.0 keep medium –
weather_compensation −23: 47, 10: 25 observed LWT setpoint 29.1 (≤ 3 °C) … 25.2 (≥ 9 °C) keep (matches); note RT modulation adds ±5 K medium –

Suggested heat_pump_efficiency (outdoor °C → COP, normalised by its maximum; values are measured daily medians, smoothed):

    heat_cop: 4.0
    heat_pump_efficiency:
      -10: 2.3
      -5: 2.5
      0: 2.75
      3: 2.85
      5: 3.0
      8: 3.3
      10: 3.3
      12: 3.6
      14: 4.0
      20: 4.0

Input entities

Setting Current entity Status Suggestion
external_temperature sensor.ecowitt_outdoor_temperature valid keep
internal_temperature sensor.bridge0_sensors_temperature_room valid keep
target_temperature sensor.bridge0_room_room_heating_setpoint valid keep
heating_energy sensor.ashp_daily_electricity valid, but includes DHW (today 0.94 kWh = 0.12 heating + 0.82 DHW) sensor.ashp_heating_power_consumption_daily (heating only, same sensor already used as car_charging_energy filter)
heating_active binary_sensor.bridge0_unknown_climate_active_q4 (as a list) verified 7 Oct: = space heating enabled, not demand; and a list is never converted to on/off, so it always reads as active heating_active: binary_sensor.bridge0_mode_valve_zone_main — single value, no -
volume_temp sensor.ashp_flow_temperature_living_room_temperature unavailable since 26 Sep (Zigbee sensor offline/battery); also ignored inside predheat: fix the sensor and move the key to the top level of pred_bat: (see note above)

Predbat interaction

  • load_forecast: predheat.heat_energy$external with car_charging_energy: sensor.ashp_heating_power_consumption_daily removes heating from historical load and adds the Predheat forecast: consistent, no double counting for space heating. DHW remains in the historical base load, which is acceptable while Predheat models space heating only.

  • Heating switched off (summer): the automation disables Predheat, predheat.heat_energy disappears and Predbat logs Warn: Unable to load the load forecast from predheat.heat_energy every cycle (harmless: the source is skipped). Predbat only warns when the series is empty (fetch.py, load_forecast → minute_data), so point it at a template sensor that passes Predheat's external list through while Predheat is enabled and otherwise publishes a flat zero series in the same format ([{last_updated, energy}], cumulative kWh, %Y-%m-%dT%H:%M:%S%z). It also ignores the stale forecast Predheat leaves behind right after it is disabled:

    # configuration.yaml
template:
  - sensor:
      - name: "Heating load forecast"
        unique_id: heating_load_forecast
        unit_of_measurement: kWh
        state: >
          {{ states('predheat.heat_energy') | float(0)
             if is_state('switch.predbat_predheat_enable', 'on') else 0 }}
        attributes:
          external: >
            {% set ext = state_attr('predheat.heat_energy', 'external') %}
            {% if is_state('switch.predbat_predheat_enable', 'on') and ext %}{{ ext }}
            {% else %}{% set t0 = today_at('00:00') %}
            [{"last_updated": "{{ t0.strftime('%Y-%m-%dT%H:%M:%S%z') }}", "energy": 0},
             {"last_updated": "{{ (t0 + timedelta(days=3)).strftime('%Y-%m-%dT%H:%M:%S%z') }}", "energy": 0}]
            {% endif %}

    and in apps.yaml: load_forecast: - sensor.heating_load_forecast$external.

  • load_ml_enable: true: confirm that Predbat's ML load model applies the same car_charging_energy filtering; otherwise space heating is counted twice (ML history + Predheat).

Next step (M1)

Replay Predheat's simulation with current vs suggested values against this winter's measurements once live data exists (or on last winter's hourly LTS), and report daily-energy and room-temperature error before/after.

List vs single value (verified in Predbat/Predheat source)

Setting Read as List allowed?
external_temperature, internal_temperature, target_temperature, heating_energy entity history (minute_data_entity) yes (single or list)
days_previous, days_previous_weight list yes (list expected)
heating_active boolean (get_arg(..., False)) no — list is always truthy
flow_temp float (sensor or number) no (single value); ignored while weather_compensation is set
volume_temp float, top level of pred_bat: only no
weather entity id no
numeric settings (heat_loss_watts, …) float/int no