GoodWe RS485 Controller

Field installation & commissioning guide — v0.1.0

Installer copy

Read this before touching anything

The inverter holds its last command forever. It has no meter-timeout. If the controller stops talking, the battery keeps charging or discharging at whatever it was last told — indefinitely.

Measured: a controller went silent mid-command and the inverter held 5 kW of discharge for 113 seconds until a person intervened.

If anything looks wrong at any point: STOP THE ADD-ON. That commands 0 W and the hardware watchdog holds it there. Nothing is at risk while you think.

Order of operations

Do not reorder. Each step assumes the previous one passed.

Step	Action	Gate to pass
0	Pre-visit checks (client, warranty, HA type)	All boxes ticked
1	Survey & photograph existing wiring	Photos taken
2	Bring system to 0 W, disconnect vendor controller	Battery idle
3	Wire and power the T-CAN485	Board online
4	Flash the ESP32 firmware	Entities appear in HA
5	Install & configure the add-on	All checks green
6	Commissioning gates 1–6	Every gate passes
7	E-stop: fit & test, or get signature	Tested or signed
8	Prove a maintenance cycle	Phases observed
9	Handover & sign-off	Sheet complete

The safety model

Layer	Covers	Where it lives
1 Watchdog	Controller alive but silent → 0 W, and keeps writing it	ESP32 firmware
2 Wind-down	Planned firmware update → 0 W before it starts	ESP32 firmware
3 E-stop	The HA machine itself dying → 0 W after 30 s of bus silence	Optional Pi + RS485

Only layer 3 covers the host dying. Nothing running on the HA machine can cover its own death. On a site without the e-stop, a failed host leaves the battery latched until someone intervenes manually. See Step 7.

Step 0 — Before you travel

Compatibility gate — 60 seconds, do this first

Read the inverter's serial number (label, app, or portal) and take characters 6–8. That three-letter tag is the platform.

Tag	Platform	Verdict
`ESU EMU ESA BPS BPU EMJ IJL`	Platform 105 — ES / EM / BP, AA55-era meter bus	SUPPORTED
`SPB SPN`	SBP G2 (GW…-SBP-20) — platform 745	NO
`ESN ESC`	ES G2 (GW…-ES-20) — platform 745	NO
`ETU EHU BTU …`	ET / EH / BT hybrids — platform 205/745/753	NO
`SDT DST MSU NSU …`	PV-only inverters — no battery to control	NO

Reference unit: 95000BPS225W0290 → BPS → supported. A “-20” or “G2” suffix in the model name means a different platform with a different meter protocol. If the tag is not on the supported list, stop — do not quote the job.

Confirm with the client

Inverter is GoodWe ES / BP family
Vendor controller may be disconnected
Warranty / installer agreement acknowledged in writing
Grid rules checked (BE: Synergrid C10/11)
Vendor subscription — cancel only after sign-off

Confirm on their system

HA is OS or Supervised (Settings → System → Repairs → ⋮ → System information)
Grid power sensor exists, updates ≤ 10 s
Note its entity id and update rate
E-stop in the van, or signature form printed

Stop conditions

If HA is Container or Core, add-ons cannot be installed and this product cannot go in. If the grid sensor updates slower than ~10 s, the default tuning is wrong for the site — do not proceed without adjusting gain.

Never do these

Never	Because
Flash third-party firmware on the inverter's WiFi dongle	Common image targets different hardware — bricks it
Long-press (3–5 s) the inverter's WiFi Reset button	Factory-resets the dongle, loses network access. Short press (~1 s) is safe
Connect our controller while the vendor box is still attached	Two masters on one bus — contradictory commands
"Compensate" for an inverted meter sign in the tuning	Fix `meter_invert` or the wiring. A wrong sign drives the grid away from zero at full gain
Raise the e-stop's 30 s threshold to stop it firing during updates	Firing during an update is correct behaviour

Steps 1–3 — Hardware

Qualified persons only. Mains AC and battery DC are present. The meter bus is low-voltage but sits inside an energised installation.

1 Survey and record

Photograph the meter-port wiring before touching it.
Identify the RS485 pair (A/B) running to the vendor controller.
Record inverter model + serial, battery capacity (Ah) and nominal voltage.

VERIFY — photos on file; A/B pair positively identified.

2 Disconnect the vendor controller

Bring the system to 0 W — battery neither charging nor discharging (check the inverter display).
Disconnect the vendor controller from the meter port.
Leave it physically mounted but unplugged — that keeps the install reversible and gives you a rollback story.

VERIFY — battery at 0 W, vendor box disconnected.

IF NOT — do not continue while the vendor box can still write to the bus.

3 Fit and power the T-CAN485

Wire A→A, B→B to the meter port, plus its own supply.
If fitting the e-stop, connect its USB-RS485 adapter to the same pair, in parallel.
Power the T-CAN485 last.

VERIFY — board joins WiFi and appears in ESPHome.

IF NOT — swapped A/B gives a completely silent bus, not an error. Swap and retry before suspecting anything else.

It transmits within seconds of boot. Never power it up while the vendor controller is connected.

Step 4 — Flash the firmware

4 ESPHome

Open the ESPHome Device Builder add-on on the client's HA.
Create a device from firmware/goodwe-master.yaml.
Set the three substitutions at the top — and nothing else.
Add the client's WiFi to ESPHome's secrets.yaml.
Install: first flash by USB, everything after that over the air.

VERIFY — device shows Online, and number.<node>_goodwe_setpoint_w exists in HA.

Substitution	Set to	Why it matters
`name`	Node name, e.g. `goodwe-master`	Write it on the sheet. Changing it later renames every entity in HA and silently breaks the add-on
`max_w`	Inverter continuous rating, e.g. `5000`	Hard firmware limit, independent of HA. Last line of defence against a controller bug
`wd_ms`	`30000`	Watchdog timeout. Must stay well above the add-on heartbeat (10 s)

ESPHome Device Builder — the node must read **Online** before you continue. (Reference bench shown.)

Testing firmware changes: an update that lands new shutdown-path code still runs the old code on the way out. If you change wind-down or watchdog behaviour, upload twice before believing a test result.

Step 5 — Install and configure the add-on

5 Add-on

Settings → Add-ons → Add-on store → ⋮ → Repositories → add the repository URL.
Install GoodWe RS485 Controller. Do not start it yet.
Open Configuration and fill in the table below.
Start the add-on, open its Web UI.
Work down the Commissioning list until every line is green.

VERIFY — banner is green or amber, no red lines.

IF NOT — the banner names the failing item. Copy entity ids from Developer Tools → States. Do not type them from memory.

Option	Set to
`meter_entity`	Client's grid power sensor. Positive must mean importing — else set `meter_invert`
`soc_entity`	`sensor.<node>_goodwe_battery_soc` — the ESP32's own read, never the cloud/dongle sensor
`batt_entity`	`sensor.<node>_goodwe_inverter_ac_power`
`setpoint_entity`	`number.<node>_goodwe_setpoint_w`
`max_w`	1000 for commissioning. Raise at the end of Step 6
`estop_fitted`	true only if you actually fitted one
`peak_forecast_entity`	Capacity-tariff sites only, else leave empty
`price_now_entity` / `price_avg_entity`	Dynamic-tariff sites only, else leave empty

Entity ids are the #1 install error

HA prefixes entity ids with the device's area at creation time, so the same firmware yields sensor.goodwe_master_… on one site and sensor.cellar_goodwe_master_… on another. A wrong entity id is not an error anywhere in HA — it simply never produces a value. On the reference system two safety alarms pointed at non-existent entities and were dead for a day while reading “on”.

Reading the Web UI

Two states you must tell apart at a glance. The banner is the summary; the Commissioning list is the detail.

Web UI running normally — **HEALTHY.** Green banner. All checks ✓. *Why* reads `tracking`. **Grid near zero** and the commanded figure **resting** between updates — a command that moves every single cycle means the deadband is too small for this meter.

Web UI showing NOT READY — **NOT READY.** Red = must fix before going further; the banner names the first failing item. Amber = permitted, but the client must have signed for it — here, no e-stop fitted. Note *Why* reads `inputs-missing` and **Commanded** has gone to **0 W**: with a source missing, the controller zeroes the inverter rather than guessing.

Step 6 — Commissioning gates

Judge every gate on the wire or the meter — never on how Home Assistant looks.

#	Do	Pass	If it fails
1	Add-on running, control stopped	Steady 0 W written, inverter idle, no errors	Check wiring and Step 5
2	By hand (Developer Tools → `number.set_value`): +300 W, then −300 W	Battery discharges ~300 W, then charges ~300 W, within ~10 s	Sign inverted → fix `batt_invert`/wiring, not the tuning
3	Compare ESP32 readings to inverter display	SoC identical; power within ~5 %	Wrong entity, or dongle sensor used by mistake
4	Set +300 W, then STOP the add-on	Inverter reaches 0 W within ~30 s and stays	STOP THE INSTALL. Nothing here is safe without this
5	`max_w`=1000, start control	Grid settles within tens of watts of zero and rests	Hunting → raise `deadband_w`; slow meter → lower `gain`
6	Switch on a kettle/oven (~2 kW)	Grid back near zero within ~20 s; command stops rising once the battery catches up	Command keeps climbing = runaway. Stop the add-on immediately

THEN — raise max_w to the value the site is sold with and repeat gate 6 once.

Step 7 — The e-stop

7a If fitted

Connect the Pi to the same RS485 pair; power it.
Run: python3 rs485_log.py --out /home/pi/bus.log --panic
Confirm it prints PANIC ARMED.
Test it: with the battery at +300 W, cut power to the T-CAN485.

VERIFY — within ~35 s the log shows a TX line and the inverter goes to 0 W.

IF NOT — check the adapter is on the correct pair and that the tap was started with --panic.

Arm the panic write only while our controller owns the bus. Pointed at a bus somebody else drives, it is unrequested interference.
It will fire during every firmware update — the upload silence exceeds 30 s. That is correct. Do not raise the threshold.

7b — If not fitted: client acknowledgement

Read this to the client and have them sign. Note it on the sign-off sheet.

Without the RS485 e-stop: if the Home Assistant machine fails, loses power, or its storage fails, the battery inverter will continue charging or discharging at whatever level it was last commanded, indefinitely, until someone intervenes manually. The inverter has no automatic fallback of its own. The e-stop is the only component that prevents this.

Client name

Signature

Date

Step 8 — Prove a maintenance cycle

8 Do not wait a month to find out it never fires

Web UI → Force maintenance cycle.
Watch the phase go drain → charge → hold → idle.
Confirm the inverter exports during drain, charges during charge, sits at 0 W during hold.
Set maintenance_enabled: true.

SHORT VERSION (~10 min) if the client cannot spare hours: set maintenance_soc_floor just below current SoC, maintenance_soc_target just below that, maintenance_hold_min: 5. Restore real values afterwards and record it as a partial test.

Why it matters: the full charge is when the BMS balances cells; the deep discharge recalibrates the coulomb counter. Skipping it breaks nothing visibly — it degrades the pack over months, and the first symptom is a state-of-charge reading nobody can trust.

Step 9 — Handover & sign-off

Tell the client, in these words

“This is the Web UI. Green means it is working.”
“If anything looks wrong, press stop. That sets the battery to zero and it idles safely.”
“Once a month it will deliberately empty and then fill the battery. That is normal and it keeps the battery healthy.”

Commissioning record

Site / client
Inverter model / serial
ESPHome node name
Add-on version / `max_w`
Meter entity + update rate

Gates 1–6 all passed on the wire or the meter
Gate 4 (watchdog) personally observed reaching 0 W
Maintenance cycle proven — full / partial (circle one)
E-stop fitted & tested OR §7b signed
Client shown the Web UI and the stop button
Vendor subscription cancelled (only after the above)

Technician

Signature

Date

Troubleshooting

Symptom	Likely cause	Do this
Web UI says NOT READY	Wrong entity id, or sensor unavailable	Banner names it. Copy the id from Developer Tools → States
Configured, but nothing moves	Control is stopped	Press start; banner turns green
Inverter does nothing, no errors	A/B swapped, or vendor box still connected	Swapped RS485 gives a silent bus, not an error
HA shows one setpoint, inverter does another	Firmware `max_w` below add-on `max_w`	Firmware wins by design. Align them
Log shows writes rejected (HTTP 400)	Value outside the number entity's range	Lower `max_w`; check the firmware substitution
Command keeps climbing while the battery is pinned	Runaway — sign inverted or saturation defeated	Stop the add-on now. Verify `meter_invert` with a known load
Grid hunts, never rests	Deadband too small, or meter too slow	Raise `deadband_w`; if meter > 10 s, lower `gain`
Battery flat overnight and stays flat	Maintenance drain never exited	Check the SoC entity is the ESP32's, not a cached cloud value
Maintenance never runs	Disabled, or not yet due	Check `maintenance_enabled` and the “Maintenance due” row
E-stop fires on every update	Expected — upload silence > 30 s	Nothing to fix. Do not raise the threshold
Add-on will not install	HA is Container/Core	Not supportable — see Step 0

Why the tuning is what it is

Every value came from hardware. Do not change them without measuring.

Setting	Value	Reason
`gain`	0.6	Inverter needs 3–6 s to settle (1.4 s dead time, 94 % by 3.3 s) against a ~5 s cycle. At the limit — do not raise
`slew_w`	1000	Most of a correction in the first cycle without running far ahead of the hardware
`deadband_w`	15	Measured residual: mean 15.4 W, max 27 W. At 10 W the command never rests
`saturation_w` / `cycles`	500 / 3	Divergence = inverter at a limit → magnitude may fall, never rise. The 3-cycle term is essential: tested instantly it fires on every large correction
`step_w`	10	Register is 1 W but response lands on a ~17.6 W ladder. Finer is meaningless
`heartbeat_s`	10	Three chances to be heard before the 30 s watchdog acts

The failure this guards against is real: the vendor controller commanded −14 547 W against an inverter reporting −5250 W, and kept climbing for six minutes, because its integrator never stopped.