Vaillant: report energy per circuit (heating vs. hot water)

[Johannes-del]

Problem

Two Vaillant chargers on the same myVaillant system (one for hot water, one with zone: for the heating circuit) report the same power to both loadpoints. charger/vaillant.go decorates the meter with the MPC sum over all devices of the system (#17994), which has no notion of circuits. Every kWh gets counted twice, and the prioritizer credits one loadpoint with the other's consumption.

A 0 W dummy meter on the heating loadpoint (the workaround suggested in #32535) stops the double counting, but then heating sits at 0 kWh forever and hot water absorbs the entire heat pump consumption:

evcc for 2026-08-20: 2.58 kWh hot water, 0 kWh heating. myVaillant app, same day: 1.6 kWh hot water plus 1.1 kWh heating.

What I'd like

myVaillant does report electrical energy per circuit. That is where the app's own split comes from: /emf/v2/{systemId}/devices/{uuid}/buckets with operationMode=HEATING|DOMESTIC_HOT_WATER and energyType=CONSUMED_ELECTRICAL_ENERGY, summed over heat pump and backup heater. sensonet already wraps it (GetEnergyData, device list via GetDeviceData), evcc just never calls it. Decorating the charger with api.MeterEnergy (HEATING when zone is configured, DOMESTIC_HOT_WATER otherwise) would give each loadpoint its real energy, since evcc prefers meter totals over power integration.

I ran this outside evcc first, as loadpoint meters fed by a small local service reading those buckets. It matches the app to within rounding. Three things I hit doing that:

• Power stays system-wide, there is no per-circuit power. Only the energy side can be fixed this way.
• Buckets are hourly and get revised downwards occasionally. The total has to be accumulated monotonically, otherwise SetEnergyMeterTotal re-anchors to the lower value and books the same energy again on the way back up.
• Standby (~12 W here) appears in no bucket, so both circuits together stay slightly below the system total. The app has the same gap.

evcc 0.314.0. Happy to send a PR if you want this.

[github-actions]

Investigated charger/vaillant.go — this matches the report.

The meter decorator added in #17994 does exactly what's described:

if devices, _ := conn.GetMpcData(systemId); len(devices) > 0 {
    implement.Has(res, implement.Meter(util.Cached(func() (float64, error) {
        res, err := conn.GetMpcData(systemId)
        return lo.SumBy(res, func(d sensonet.MpcDevice) float64 {
            return d.CurrentPower
        }), err
    }, cc.Cache)))
}

It sums CurrentPower over the whole MPC device tree of the myVaillant system, with no concept of circuits (heating zone vs. DHW). When two vaillant chargers are configured against the same system, both get the identical system-wide value, so the same consumption is booked twice. This is a design limitation from when power reading was first added, not a regression — there is no per-circuit split anywhere in the current code.

The proposed fix (decorating with api.MeterEnergy fed from sensonet's per-circuit /emf/.../buckets energy data, split by operationMode) sounds like a reasonable way to give each loadpoint its own energy total, given evcc prefers meter totals over power integration when available. I can't verify the exact sensonet wrapper API surface (GetEnergyData/GetDeviceData) from here, and the monotonic-accumulation handling for revised/backdated buckets adds real complexity, so this needs implementation against real hardware/API responses to get right — not something to attempt blind.

Marking as heating since this is heat-pump/circuit specific. Happy to see a PR for this given you've already validated the approach externally.

🤖 Generated with Claude Code

[Johannes-del]

Both open points are covered.

sensonet (v0.0.7, the version in go.mod) already has what is needed, and evcc references neither function today:

• GetEnergyData(systemId, deviceUuid, operationMode, energyType, resolution, startDate, endDate) in connection.go:165, on ENERGY_URL = /emf/v2/%s/devices/%s/buckets?. The response carries TotalConsumption, so one query per circuit is enough.
• GetDeviceData(systemId, whichDevices) in connection.go:136, on DEVICES_URL = /emf/v2/%s/currentSystem, for the device UUIDs.

One catch: currentSystem lists more than one emf device. Here it returns the aroTHERM plus (HEATPUMP) and the hydraulic station (ELECTRIC_AUXILIARY_HEATER), both exposing CONSUMED_ELECTRICAL_ENERGY for HEATING and DOMESTIC_HOT_WATER, and both have to be summed. For 2026-08-20 that gives 1.60 kWh DHW and 1.07 kWh heating against the app's 1.6 and 1.1.

The monotonic accumulation is less work than it looks: poll HOUR buckets from local midnight, keep a persistent counter, add only max(0, sum - lastSum), and on a day rollover pull the previous day's final sum once before starting the new day. It does matter though, since SetEnergyMeterTotal skips a decrease but re-anchors the baseline to it, so a bucket revised downwards would be booked twice on the way back up.

Two details for an implementation: DAY buckets use UTC boundaries while HOUR buckets come back in local time, so HOUR is what matches the app's day, and the current hour is already populated, so a 5 minute poll is plenty.

Running as loadpoint meters here since yesterday, both loadpoints report charge: power ✓ energy ✓ with totals matching the app.

[andig]

A logfile or PR would help. Somehow this needs a device selector.