Skip to content

Sensors

One integration instance creates three sensors, all grouped under a single device.

Overview

Sensor Unit Device Class State Class
Time remaining h — Measurement
Rate of change <unit>/h — Measurement
Predicted time — timestamp —

Time remaining

Shows how many hours it will take at the current trend to reach the active target (minimum or maximum, depending on direction).

  • Unit: h (hours, as decimal — e.g. 4.2)
  • To display minutes in a dashboard: {{ (states('sensor.xyz_time_remaining') | float * 60) | round }} min
  • Shows unavailable when:
    • Fewer than 2 data points in the time window
    • The source sensor does not provide a numeric value
    • The rate is exactly zero (flat trend)
  • Attribute target: which value is currently being predicted towards (min or max)

Rate of change

The current rate of change of the source sensor per hour, calculated via linear regression.

  • Unit: dynamic — %/h if the source sensor uses %, otherwise <unit>/h
  • Negative = value is falling → integration predicts time to minimum
  • Positive = value is rising → integration predicts time to maximum

Predicted time

The concrete point in time at which the active target is expected to be reached — as a timestamp sensor.

  • Device class: timestamp → HA formats it automatically as date/time
  • Can be used directly as a time trigger in automations
  • Shows unavailable under the same conditions as the time remaining sensor
  • Attribute target: which value is currently being predicted towards

How direction detection works

The integration picks the target automatically based on the regression slope:

rate < 0  →  target = min_value  (e.g. 0)
rate > 0  →  target = max_value  (e.g. 100)

time remaining = (target − current value) / rate

When a PV battery switches from discharging to charging, the rate sign flips and both the target and the predicted time update automatically — no reconfiguration needed.

How the calculation works

The integration collects all state changes of the source sensor within the configured time window and calculates a trend line using linear regression (least squares). The time remaining is derived from the current value, the slope, and the active target.

Linear regression over multiple points is more robust than a simple first/last difference because short-term outliers carry less weight.