Cell balance monitor¶
Tracks the voltage spread between the highest and lowest cell at the top of a full charge. The reading is used to show whether the battery pack is staying balanced over time and to trigger imbalance alerts when the spread becomes high.
Why this is needed on LFP batteries¶
Marstek Venus batteries use LFP cells. LFP is very stable and long-lived, but its voltage curve is almost flat through most of the usable SOC range. Around the middle of the charge, two cells can have noticeably different SOC while still reporting almost the same voltage. That makes mid-SOC voltage readings poor indicators of cell balance.
The useful balance window is near the top of charge. Above roughly 3.45 V per cell, the LFP voltage curve rises much faster, so differences between cells become visible. That is also the area where the battery BMS is expected to perform passive balancing by bleeding the highest cells.
In practice, the Marstek BMS does not always balance the cells well by itself. If the pack reaches 100% quickly and then immediately returns to normal operation, weak balancing can leave one cell consistently higher than the rest. This integration therefore does two things:
- it slows the final part of a 100% charge so the BMS has time to work in the top-balance window;
- it measures imbalance only at a repeatable top-voltage point, instead of using noisy mid-SOC readings.
The LFP charge curve in detail¶
LFP (LiFePO4) chemistry has a charge/discharge curve that is fundamentally different from Li-ion NMC or NCA. Understanding it is what justifies every voltage threshold this integration uses.
A typical 3.2 V nominal LFP cell behaves like this during a constant-current charge:
| SOC range | Cell voltage range | Slope |
|---|---|---|
| 0 – 10 % | 2.50 V → 3.20 V | Steep entry knee |
| 10 – 90 % | 3.20 V → 3.30 V | Almost flat — about 1 mV per % SOC |
| 90 – 97 % | 3.30 V → 3.45 V | Mild rise begins |
| 97 – 99 % | 3.45 V → 3.55 V | Knee — voltage starts climbing sharply |
| 99 – 100 % | 3.55 V → 3.65 V | Steep top knee — full-charge cliff |
That long flat plateau is the reason LFP voltage tells you almost nothing about state of charge in the middle of the curve. Two cells that look identical at 3.28 V can in reality have 5 – 10 % SOC of difference between them, which is huge.
The plateau also means the BMS cannot do meaningful passive balancing in the middle of the curve. Passive balancing works by bleeding current off the highest cell through a resistor. To even detect which cell is "highest", the BMS needs the spread between cells to rise above measurement noise. On the plateau, all cells read essentially the same number, so the BMS has nothing to act on.
Only when the pack enters the upper knee (above roughly 3.45 V) do cell voltages spread apart enough for the BMS to identify the leader. A 10 mV spread on the plateau might correspond to a 5 % SOC difference, but the same 10 mV spread above 3.50 V represents a tiny SOC delta — which is exactly what you want at the end of charge.
So balancing on LFP is only effective in a narrow window: roughly the last 1 – 3 % of charge, above 3.45 V. Outside that window the BMS is essentially blind to imbalance, and any time the pack spends below the knee is time the cells are not getting balanced.
Availability¶
The cell balance monitor is always active. There is no separate configuration option because the readings are useful battery health data and do not change normal operation by themselves.
There are two related controls that decide when the battery is taken to the top-voltage measurement window:
- 100% charge voltage taper: per battery option. When the charge target is 100%, the integration slows the final charge and records a top-voltage balance reading.
- Active balance mode: per battery switch. When enabled, the integration actively cycles that battery at the top until the measured cell delta is low enough.
The weekly full charge feature can temporarily set the battery max SOC to 100%. Once it does that, the same 100% charge voltage taper rules are used.
100% charge voltage taper¶
This path is used when a battery has a 100% charge target:
- the user configured that battery with
max_soc = 100, or - the weekly full charge temporarily raised the battery to 100%.
The weekly full charge does not use a different balance profile. It only changes the target SOC to 100%; voltage thresholds, charge power and measurement logic are the same.
Charge profile¶
flowchart TD
A([Normal charging]) --> B(Charge with configured charge limit)
B --> C{max_cell_voltage < 3.48 V}
C -->|Yes| A
C -->|No| D([Limited charging])
D --> E(Charge with 95 W)
E --> F{max_cell_voltage < 3.58 V}
F -->|Yes| D
F -->|No| G(Stop charging and wait 60 s)
G --> H("Record cell_delta = (cell_Vmax - cell_Vmin) * 1000")
H --> I([Normal SOC charge/discharge logic])| Condition for one battery | Action |
|---|---|
max_cell_voltage below 3.48 V |
Normal configured charge limit |
max_cell_voltage at or above 3.48 V |
Limit charge to 95 W |
max_cell_voltage at or above 3.58 V |
Stop charging and wait 60 s |
| After the 60 s wait | Record delta_mV = (Vmax - Vmin) * 1000 |
The logic is voltage based. SOC is deliberately not used to decide when the top-voltage taper starts or stops, because SOC can be less reliable near the top than the cell-voltage registers.
There is no extra voltage hysteresis in this path. Once the battery reaches 3.58 V and the reading has been taken, the integration does not force a discharge. It leaves charging stopped at that voltage and lets the normal SOC/charge logic decide when future charging is allowed again.
In a multi-battery system, this is evaluated per battery. One battery can be limited or paused while another continues charging normally.
Active balance mode¶
Active balance mode is a stronger per-battery recovery mode for packs that need more time in the balancing window.
When the switch is enabled, that battery is excluded from normal PD control. The rest of the batteries can continue to operate normally. The integration temporarily raises the battery charge target to 100% and commands charge directly for that battery.
Active balance profile¶
| Phase | Action |
|---|---|
| Before the top window | Charge from the grid at the battery's configured maximum charge power until max_cell_voltage >= 3.49 V |
| Regulated top charge | Charge at 95 W until max_cell_voltage >= 3.58 V |
| Measurement wait | Stop charge/discharge, wait 60 s, then measure cell delta |
If delta_V > 0.03 V |
Discharge at 25 W until max_cell_voltage <= 3.49 V, then charge again |
If delta_V <= 0.03 V |
Final discharge at 25 W until max_cell_voltage <= 3.48 V, then finish and turn the switch off |
If the BMS accepts the charge command but does not actually charge before reaching 3.58 V, the integration treats that as charge rejection. It discharges first and lowers the retry voltage by 0.01 V, down to a minimum of 3.40 V, so the next cycle can restart from a point where the BMS is more likely to accept charge.
Active balance mode has no fixed 48-hour timeout. It runs until the measured top-voltage delta is at or below 0.03 V, or until the user turns the switch off.
Why these voltage thresholds¶
Every voltage cutoff used by the 100 % taper and the active balance mode was picked against the LFP curve described above. None of these numbers are arbitrary.
| Threshold | Where it is used | Why this value |
|---|---|---|
| 3.45 V | Reference for the start of the upper knee | This is roughly where the LFP curve leaves the plateau. Below this, balancing decisions cannot be trusted because cell voltages are too close together to distinguish. |
| 3.48 V | Trigger for tapering the charge to 95 W | A little above the knee. The small margin confirms the pack is genuinely in the balance window — and not just on a brief voltage bounce caused by a load step — before reducing power. |
| 3.49 V | Discharge floor between active-balance retries; switch-over from coarse to regulated charge | Sits just inside the balance window. Stopping the discharge here keeps the pack in the zone where the BMS can still see and bleed the high cell. Going lower would push the pack off the knee and waste the time already spent balancing. |
| 3.58 V | Top measurement point; stop charge and wait 60 s before reading the delta | High enough that even the lowest cell is firmly in the knee, so the spread between cells is meaningful. Low enough that the highest cell is still safely below the 3.65 V LFP datasheet ceiling and the BMS over-voltage cutoff. The ~70 mV headroom is intentional: the spread between cells is what we are trying to measure, and you must leave room for it. |
| 3.48 V (again) | End-of-cycle discharge floor — the 25 W final discharge after a completed active-balance run stops here | The same threshold used to enter the taper is reused to leave the balance window. Stopping at 3.48 V brings the pack just off the upper knee without dropping it back onto the deep plateau. Sitting at 3.55 – 3.58 V for long periods accelerates calendar ageing, so the integration deliberately bleeds the pack down to the lower edge of the window before releasing control. |
| 3.40 V | Lower bound for the active-balance retry voltage when charge rejection is detected | The integration drops the retry voltage by 0.01 V each time the BMS rejects charge, but never below 3.40 V. Going further down exits the balance window entirely and forces a long, wasteful re-climb up the curve. |
| 0.03 V (30 mV) | Active-balance completion threshold | Considered "balanced enough" for an LFP pack at the top of the knee. Pushing for tighter values (10 mV or less) is rarely productive because passive balancing currents are tiny — see the next section. |
| 0.05 V (50 mV) | Green / yellow status boundary | A pack reading below 50 mV at the top is considered healthy. This is more conservative than typical LFP vendor specs (often 80 – 100 mV) because the measurement is taken in the balance window, where differences between cells are exaggerated. |
The 95 W and 25 W power values are paired with these thresholds intentionally. They are low enough that the cell voltage measured under load is dominated by the cell's actual chemistry rather than by IR (resistive) drop across the cell, busbars and BMS shunts. Charging or discharging at hundreds of watts in the knee would shift the apparent reading by tens of millivolts and ruin both the threshold checks and the final delta measurement.
Why this takes so long¶
Cell balancing is not a fast process — and Marstek Venus packs are no exception. There are two reasons.
1. Passive balancing current is small. A typical LFP BMS bleeds the highest cell through a balance resistor at somewhere between 30 mA and 150 mA. The Marstek Venus packs sit at the low end of that range. For a 100 Ah cell, a 50 mA bleed removes only about 0.05 % SOC per hour from the high cell. Equalising even small SOC differences between cells therefore requires many hours of continuous time in the balance window.
2. The balance window itself is narrow. The BMS can only bleed when the pack is above ~3.45 V and the highest cell is detectably above the rest. As soon as charging stops or the pack drops back below the knee, balancing stops. A normal charge cycle that hits 100 % and immediately returns to discharge spends only minutes in the useful window — far too little for any visible effect.
The practical consequence:
Reducing the top-of-charge cell delta by roughly 5 mV typically takes around 24 hours of cumulative time at the top of the balance window.
That figure is consistent both with the bleed-current arithmetic above and with observations on real Venus packs. Bigger imbalances (50 mV or more) can take multiple days of repeated top-balance sessions before the delta starts dropping consistently. Packs that have been left chronically unbalanced for months may take a week or more to recover.
This is also why active balance mode does not have a "fast" path:
- the 95 W charge cap above 3.48 V is set so the pack stays in the knee long enough for the BMS to make progress, rather than ramming through it in seconds;
- the 25 W discharge between retries is set so the pack stays near 3.49 V without dropping out of the window;
- the active-balance loop is allowed to run indefinitely, because anything short of "many hours" is unlikely to move the needle.
If the goal is to restore a noticeably unbalanced pack, the right approach is to enable active balance mode and leave it running overnight (or longer) and check the result the next day. Watching the cell delta in real time and expecting movement within minutes will only cause frustration.
How imbalance is measured¶
The only reading that feeds the balance status, alerts and trend is the explicit top-voltage measurement:
- the battery reaches
max_cell_voltage >= 3.58 V; - charge is stopped;
- the integration waits 60 seconds;
- it records the spread between
max_cell_voltageandmin_cell_voltage.
Older OCV-style readings, opportunistic readings and long passive-hold readings are no longer used. Measuring at the same top-voltage point makes readings more comparable from one full charge to the next.
Thresholds¶
| Status | Delta range | Meaning |
|---|---|---|
| Green | < 50 mV | Good balance |
| Yellow | 50-99 mV | Minor imbalance; monitor over time |
| Orange | 100-149 mV | Moderate imbalance |
| Red | >= 150 mV | High imbalance |
Thresholds are fixed and apply equally to all supported LFP packs.
Notifications¶
The integration sends Home Assistant persistent notifications for these events:
| Event | Notification title |
|---|---|
| Orange or red top-voltage reading | Cell imbalance - {battery name} |
| Red on 2 or more consecutive full charges | Possible degraded cell - {battery name} |
| Rising trend with average above 75 mV | Rising imbalance trend - {battery name} |
| Active balance mode start/finish | Active balancing started/finished - {battery name} |
Sensor entities¶
Five sensor entities are created per battery when the feature is enabled:
| Entity | Description | Unit |
|---|---|---|
sensor.*_cell_delta |
Voltage spread between max and min cell | mV |
sensor.*_balance_status |
Balance result: green / yellow / orange / red |
- |
sensor.*_delta_trend |
Trend over recent readings: rising / stable / falling |
- |
sensor.*_last_balance_read |
Timestamp of the last reading | timestamp |
sensor.*_delta_avg_4w |
Rolling average of the last 4 readings | mV |
Values are restored from persistent storage after a Home Assistant restart so sensors show the last known state immediately on startup.
Diagnostics¶
The Integration Status sensor exposes a normal_balance_protection attribute with per-battery details:
| Attribute | Meaning |
|---|---|
enabled |
Whether 100% voltage taper is enabled for that battery |
in_zone |
Whether max_cell_voltage is in the top-balance window |
paused |
Whether charging is currently stopped by high cell voltage |
max_cell_voltage / min_cell_voltage |
Current cell voltage extremes |
delta_V |
Current voltage spread in volts |
voltage_taper_latched |
Whether the 95 W taper is currently active |
active_balance_phase |
Current 100% top-measurement phase, if any |
charge_limit_w |
Effective per-battery charge limit before allocation |
Active balance mode also exposes its current phase, measured delta, command power and retry voltage through the integration status diagnostics.
Info
Cell voltage registers (max_cell_voltage, min_cell_voltage) are read from all supported battery versions (v2, v3, vA, vD).