[driver] Field run: SolarEdge SE17K (legacy) + Pixii Home 20 kWh — telemetry and control evidence #175

Device make and model

SolarEdge SE17K (fw 3.2537) + Pixii Home 20 kWh — drivers solaredge_legacy v0.3.3 and pixii v2.1.8

Protocol or local API

Modbus TCP (SunSpec on SolarEdge, port 502; Pixii local Modbus, also used as site meter)

Public protocol documentation

No response

Hardware available for testing

Yes — I own and run both devices on a live site and can run specific tests on request.

Site

• Inverter: SolarEdge SE17K, firmware 3.2537, 37 panels / 11.1 kWp
• Battery: Pixii Home 20 kWh, local Modbus TCP, also used as site meter
• Charger: none
• Main fuse: 3x20 A (Sweden)
• FTW: native v0.140.3-beta.1 (optimizer 0.5.1) on Lubuntu 26.04.1, amd64, bridged to Home Assistant via MQTT
• Drivers: solaredge_legacy v0.3.3, pixii v2.1.8 (both shown as "this release · untested")
• Running this hardware with FTW since spring 2026 (HA add-on first, native since Oct 2026)

Telemetry (checked against the vendors' own apps, 2026-10-10)

• SolarEdge PV power matches the SolarEdge app (50 W vs 0.05 kW)
• Pixii SoC: FTW shows 90 %, the Pixii app shows 100 %. Pixii presents its own charge ceiling as 100 %; the raw register value is 90.
• Pixii battery power: app shows "Idle", FTW −21 W (standby draw, plausible)
• Pixii as site meter: grid, per-phase V/A/W and frequency are reported; 0 consecutive errors over the session

Commands with verified physical effect

• Pixii discharge: measured −5.1 kW (at the configured 5 kW max discharge)
• Pixii charge: measured +2.5 to +2.9 kW
• Pixii idle setpoint (0 W) is followed
• SolarEdge curtail: not available, see below

Finding: charge requests at Pixii's SoC ceiling are silently not delivered

From 11:41 the planner repeatedly sent charge setpoints of roughly 480–885 W (register 39905) while SoC was 90 %. Measured battery power stayed at −16 to −21 W. Pixii stops charging at its own ceiling (90 % raw), but FTW has no max-SoC limit configured and plans as if 100 % is reachable. The planner therefore books energy that the device will never accept. Ideally the driver would expose Pixii's effective ceiling, or Core would learn it from the refused requests.

Not tested / unverified

• SolarEdge curtailment: never observed. The driver logs nominal_w not set in config — curtail action will be unavailable, so on this site the declared curtail capability depends on a config value that the setup flow never asked for.

Known issues on this hardware

• SE17K fw 3.2537: holding registers 40069–40120 used to fail as a single PDU read. On v0.140.3 they read fine unchunked (4 608 ticks, 0 errors).
• MPPT block (40123) is not readable on this firmware. The driver now detects this and emits Model 103 only. Note: the support dump still contains pv_mppt1/2_v/a series with constant 0, which could be mistaken for real measurements.
• At startup: release APC enable+limit failed … connection refused. This happened once, during the Modbus reconnect, and is harmless as far as I can tell.
• Pixii must be in demand response mode for external Modbus control; "adaptive peak shaving" conflicts.
• A second Modbus client (an old HA YAML integration) caused transaction ID collisions; FTW should be the only Modbus master.

Relates to the "Mixed equipment" row in srcfl/ftw docs/roadmap.md. Support dump available on request.

Contribution

• I can provide device documentation or help test on hardware.