EG4 18kPV hybrid inverter

Register map(s) for this device, each at the trust rung it earned. Addresses are 0-based as on the wire; word order and scaling are only confirmed at the hardware-verified rung.

interop-verified

Read + decoded cleanly over a real network vs an independent third-party Modbus test server (transport and framing; reads are remapped into the server's register window, so the map's own addresses are NOT exercised). Does NOT prove word order/scaling against the actual device.

point	address (0-based)	type	scale	unit
state	0	U16	1
vpv1	1	U16	0.1	0.1V
vpv2	2	U16	0.1	0.1V
vpv3	3	U16	0.1	0.1V
vbat	4	U16	0.1	0.1V
soc	5	U16	1	%
ppv1	7	U16	1	W
ppv2	8	U16	1	W
ppv3	9	U16	1	W
pcharge	10	U16	1	W
pdischarge	11	U16	1	W
vac_r	12	U16	0.1	0.1V
vac_s	13	U16	0.1	0.1V
vac_t	14	U16	0.1	0.1V
fac	15	U16	0.01	0.01Hz
pinv	16	U16	1	W
prec	17	U16	1	W
iinv_rms	18	U16	0.01	0.01A
pf	19	I16	0.001	0.001
veps_r	20	U16	0.1	0.1V
veps_s	21	U16	0.1	0.1V
veps_t	22	U16	0.1	0.1V
feps	23	U16	0.01	0.01Hz
peps	24	U16	1	W
seps	25	U16	1	VA
ptogrid	26	U16	1	W
ptouser	27	U16	1	W
epv1_day	28	U16	0.1	0.1kWh
epv2_day	29	U16	0.1	0.1kWh
epv3_day	30	U16	0.1	0.1kWh
einv_day	31	U16	0.1	0.1kWh
erec_day	32	U16	0.1	0.1kWh
echg_day	33	U16	0.1	0.1kWh
edischg_day	34	U16	0.1	0.1kWh
eeps_day	35	U16	0.1	0.1kWh
etogrid_day	36	U16	0.1	0.1kWh
etouser_day	37	U16	0.1	0.1kWh
vbus1	38	U16	0.1	0.1V
vbus2	39	U16	0.1	0.1V
epv1_all	40	U32LE	0.1	0.1kWh
epv2_all	42	U32LE	0.1	0.1kWh
epv3_all	44	U32LE	0.1	0.1kWh
einv_all	46	U32LE	0.1	0.1kWh
erec_all	48	U32LE	0.1	0.1kWh
echg_all	50	U32LE	0.1	0.1kWh
edischg_all	52	U32LE	0.1	0.1kWh
eeps_all	54	U32LE	0.1	0.1kWh
etogrid_all	56	U32LE	0.1	0.1kWh
etouser_all	58	U32LE	0.1	0.1kWh
tinner	64	I16	1	℃
tradiator1	65	I16	1	℃
tradiator2	66	I16	1	℃
tbat	67	I16	1	℃
max_chg_curr	81	U16	0.01	0.01A
max_dischg_curr	82	U16	0.01	0.01A
charge_volt_ref	83	U16	0.1	0.1V
dischg_cut_volt	84	U16	0.1	0.1V
bat_capacity	97	U16	1	Ah
bat_current_bms	98	I16	0.01	0.01A
max_cell_volt_bms	101	U16	0.001	0.001V
min_cell_volt_bms	102	U16	0.001	0.001V
max_cell_temp_bms	103	I16	0.1	0.1℃
min_cell_temp_bms	104	I16	0.1	0.1℃
bat_volt_sample_inv	107	U16	0.1	0.1V
t1	108	I16	0.1	0.1℃
v_bus_p	120	U16	0.1	0.1V
gen_volt	121	U16	0.1	0.1V
gen_freq	122	U16	0.01	0.01Hz
gen_power	123	U16	1	W
egen_day	124	U16	0.1	0.1kWh
egen_all	125	U32LE	0.1	0.1kWh
eps_volt_l1n	127	U16	0.1	0.1V
eps_volt_l2n	128	U16	0.1	0.1V
peps_l1n	129	U16	1	W
peps_l2n	130	U16	1	W
seps_l1n	131	U16	1	VA
seps_l2n	132	U16	1	VA
eeps_l1n_day	133	U16	0.1	0.1kWh
eeps_l2n_day	134	U16	0.1	0.1kWh
eeps_l1n_all	135	U32LE	0.1	0.1kWh
eeps_l2n_all	137	U32LE	0.1	0.1kWh

link settings as documented (unverified): RTU; default 19200 8N1; FC 03,04,06,16; Minimum polling period: 1s. Register width: 2 bytes.

bench facts as documented (unverified)

terminals	RS-485
wiring notes	Communication method: Universal Asynchronous Transceiver (UART)
max registers/read	40
protocol notes	The inverter software groups the registers into 40 groups of 0-39, 40-79, 80-119 and so on. When querying 40 registers, the starting address must be 0 (0-39), 40 (40-79 ), 80 (80-119), it is not allowed to query registers across groups at the same time.
serial register	Hold 2-6
firmware register	Hold 9-10

field reports — community-sourced, unverified (checked 2026-06-11)

Register 67 (Tbat) is documented as battery temperature in degrees C, but in practice, it does not return battery temperature; instead, it acts as a bitfield or counter for battery status / Over-Current Protection (OCP) count. [source]
The requirement to include the inverter's serial number in every Modbus request applies to the INV485 port (shared with the Wi-Fi module), but does not apply when communicating via the METER485 port. [source]
There are significant discrepancies in bit assignments for registers like 110 (H_FUNCTION_ENABLE_3) and 179 (H_FUNCTION_ENABLE_4) between the official EG4 spec and actual Luxpower-based code implementations. [source]
confirmed: comms defaults (baud rate 19200, no parity, 1 stop bit, RTU framing, function codes 3, 4, 6, 16) — confirmed by the official EG4 18KPV 12LV Modbus Protocol document [3.1.1]
confirmed: quirks: max_read_registers: 40, and register grouping (0-39, 40-79, 80-119, etc.) — confirmed by the official EG4 18KPV 12LV Modbus Protocol document
confirmed: identification: firmware_register: Hold 9-10, serial_register: Hold 2-6 — confirmed by the official EG4 18KPV 12LV Modbus Protocol document
confirmed: register map summary (state @ 0, vpv1 @ 1, vpv2 @ 2, vpv3 @ 3, vbat @ 4, soc @ 5, ppv1 @ 7, ppv2 @ 8, ppv3 @ 9, pcharge @ 10, pdischarge @ 11, vac_r @ 12, vac_s @ 13, vac_t @ 14, fac @ 15, pinv @ 16, prec @ 17, iinv_rms @ 18, pf @ 19, veps_r @ 20, veps_s @ 21, veps_t @ 22, feps @ 23, peps @ 24, seps @ 25, ptogrid @ 26, ptouser @ 27, epv1_day @ 28, epv2_day @ 29, epv3_day @ 30, einv_day @ 31, erec_day @ 32, echg_day @ 33, edischg_day @ 34, eeps_day @ 35, etogrid_day @ 36, etouser_day @ 37, vbus1 @ 38, vbus2 @ 39, epv1_all @ 40) — confirmed by the official EG4 18KPV 12LV Modbus Protocol document and community integrations

generated by gemini:gemini-3.5-flash · harvested from https://www.dth.net/solar/luxpower/modbus/EG4-18KPV-12LV-Modbus-Protocol.pdf (EG4 Electronics); manual-faithful — the document's claims, panel risk: LOW — panel concurs, no field disagreements; proved interop-verified on 2026-06-10 · recorded 2026-06-10

The trust ladder

Each rung states what it proved and what it did not. Gray until proven; green is earned by hardware only.

generated — A prompt produced it. Unproven — anyone can regenerate this.
structure-verified — Decode path is structurally sound (addresses, widths, finite values). Does NOT prove word order or scaling.
interop-verified — Read + decoded cleanly over a real network vs an independent third-party Modbus test server (transport and framing; reads are remapped into the server's register window, so the map's own addresses are NOT exercised). Does NOT prove word order/scaling against the actual device.
simulator-verified — Decoded correctly against a protocol emulator seeded with known-true values — catches word-order/scaling errors vs the emulator, not the physical device.
hardware-verified x1 — Matched a real device with operator-attested readings as ground truth — the only rung that confirms word order + scaling on the wire.

Have a device that isn't here? Upload its register map — generation is free, and the connector earns its rung by being proven, not claimed.

Need it proven on your own device? Hardware verification — if it doesn't verify, you don't pay.

Building a platform? License the verified library — machine-readable maps, firmware matrix, API access.