Battery rate
The battery_rate strategy estimates charging power from changes in a battery percentage sensor.
It is useful when a charger changes speed but the device does not expose its charging mode.
This strategy is available in YAML and library profiles; there is no manual configuration form yet.
Power is calculated as:
power [W] = base_power + (battery_now - lowest_battery_in_window) * 3600/window * energy_per_percent
The battery percentage is held constant between readings. window is measured in seconds and
energy_per_percent in Wh per displayed percentage point. The coefficient must be measured for
the device; it is not the battery's nominal capacity divided by 100, because charging losses and
the device's displayed percentage also affect it. Battery ageing or replacement can change the coefficient.
Configuration
| Option | Required | Default | Description |
|---|---|---|---|
entity_id |
Yes | Battery percentage entity, reporting a number from 0 to 100. | |
energy_per_percent |
Yes | Positive charging energy per displayed percentage point, in Wh/%. | |
base_power |
No | 0 |
Non-negative baseline power in W, included in the result. |
window |
No | 600 |
Positive smoothing window in seconds. |
max_power |
No | No limit | Maximum total power of this strategy, including base_power, in W. Must be at least base_power. |
For example, while the battery gains 5 percentage points over 10 minutes, a coefficient of
0.96 Wh/% and baseline of 3.5 W produce 3.5 + 5 * 6 * 0.96 = 32.3 W.
The following values are illustrative and require calibration for your device.
powercalc:
sensors:
- entity_id: vacuum.robot
composite:
- condition:
condition: state
state: docked
battery_rate:
entity_id: sensor.robot_battery
energy_per_percent: 0.96
base_power: 3.5
window: 600
max_power: 60
- fixed:
power: 0
Use the device's docked or charging status for the condition. Avoid a condition such as
battery < 100: the delayed contribution after the last percentage increment helps account for
the energy used earlier in the charge. Once the percentage stops changing, the contribution above
base_power reaches zero within one window. Powercalc schedules a recalculation when each
percentage increment leaves the window, with a periodic update at least every 60 seconds
(or every window seconds for shorter windows) as a fallback.
No derivative helper or separate power_update_interval setting is needed.
With stop_at_first (the default), put wash and auto-empty branches before the charging branch,
because a docked device can perform these operations too. Drying can happen while charging;
use sum_all to add its independent power, or separate charging branches with the corresponding
base_power for drying on and off. For Roborock devices whose status remains charging at 100%,
that status can be used directly as the charging condition.
A battery_rate block can also be used directly on a sensor. It then applies whenever the parent
source is active; add a composite condition when charging needs to be distinguished from other states.
Library profiles
The same configuration works in a composite profile, including related entity placeholders:
{
"calculation_strategy": "composite",
"composite_config": [
{
"condition": {"condition": "state", "state": "docked"},
"battery_rate": {
"entity_id": "[[entity_by_device_class:battery]]",
"energy_per_percent": 0.96,
"base_power": 3.5,
"window": 600,
"max_power": 60
}
},
{"fixed": {"power": 0}}
]
}
A standalone profile uses calculation_strategy: battery_rate with battery_rate_config.
Use composite sum_all to add independent contributions such as mop drying. Keep that contribution
out of base_power and the measured charging coefficient to avoid counting it twice. max_power
limits only its own strategy's contribution, not the composite total.
Startup, interruptions and accuracy
After setup, reload or restart, the first valid reading supplies the starting percentage. There is no Recorder history lookup or restored rate. The full window remains the denominator during startup, so the estimate initially runs low rather than assuming the battery started at 0%.
Missing, unknown, unavailable, non-numeric, non-finite and out-of-range battery readings are ignored for up to one window since the last valid reading. During a short gap, recent increments still expire on schedule. A gap lasting one window discards the history and yields no calculation; composite then follows its normal branch and fallback rules. Recovery after a long gap, or with a lower percentage after an invalid reading, starts a new window without counting a charging jump.
Battery readings continue to populate the window while another composite branch is selected. When the charging condition becomes false, its power contribution stops immediately. No pending energy is added or compensated. A quick return to the charging branch reuses the recent valid history. The lowest percentage in the window supplies the reference, so a higher reading from before docking does not hide a new charge. A restart or a long invalid gap clears the history.
The smoothing window delays changes in estimated power. A longer window reduces fluctuations from whole-percentage steps but increases startup lag and energy lost when a charge is interrupted. A short charge can therefore have substantial relative energy error. Clamping large jumps can also reduce counted energy. Assess total energy and instantaneous power separately when calibrating a profile, and validate on charges that were not used to choose the coefficient.
See the draft replay evaluation for measured energy and power errors.