Solis: cap slot currents at the inverter rating and distrust implausible recovery SOC

[springfall2008]

This is an automated draft PR generated from issue #5187 — a maintainer should review it before merging.

Fixes #5187

Summary

Slot current capped at the inverter's rating. CID 7224/7226 are battery limits, not inverter ones. On the affected 3.6 kW unit they read 100 A (4.8 kW at 48 V), and the inverter refused every 100 A discharge-slot write. That left the V2 slot at 0 A, holding the battery through each export window. The same model reading 70 A or 62.5 A accepts its writes.

A new get_rated_current() works out the rating as inverterDetail power (the inverter_size sensor, already bound to inverter_limit) divided by get_nominal_voltage(), floored to whole amps. On a 3.6 kW unit that is 75 A at 48 V and 70 A at 51.2 V. The cap is applied at the point the current is written:

• V2: the rating joins the existing min(CID 7226, sysCommand.max) limit in write_time_windows_if_changed().
• V1: the CID 103 encoder caps each current as it formats it. V1 had no current limit at all before this.

Implausible Recovery SoC. 19 of 29 inverters in the fleet report CID 7229 as 0, 1, equal to the over-discharge SoC, or 65521. Any of those disabled the #4706 clamp, so a cut-off at the over-discharge SoC was written, refused, and the stale 40–50 % stayed. resolve_discharge_soc_floor() now treats a reading at or below the over-discharge SoC, or above 100, as over-discharge + 1, which is where #4702 found the inverter holds it. That fallback is never written back to CID 7229.

Testing

• ./run_pre_commit: every hook passed, and the quick suite it runs passed.

• ./run_all --test solis: passes with the fix and fails without it (recovery reads 0: discharge SOC should be 13, got 12).

• The module run stops at its first failure, so each new test was also run on its own against the unfixed solis.py:

  Test	Without the fix	With the fix
  test_implausible_recovery_soc_falls_back_to_inverter_minimum	fail	pass
  test_get_rated_current	fail	pass
  test_v2_slot_currents_capped_at_inverter_rating	fail (writes 100.0)	pass (75.0)
  test_v1_slot_currents_capped_at_inverter_rating	fail (encodes 100,100,…)	pass (75,75,…)
  test_slot_currents_uncapped_when_inverter_size_unknown	pass	pass

  The last row passes on both on purpose. It checks that an inverter with no reported size behaves exactly as before.

Notes

• Not done: the planner's own rate. The issue also asks for the same cap on battery_rate_max / discharge rate. That is left out deliberately. battery_rate_max is still CID 7224 × V (4800 W), so Predbat still asks for 4800 W and the slot power entity keeps showing it while 75 A goes to the inverter. This is the same entity/register mismatch the debug journal records for the sysCommand.max clamp (GH#5068). The plan is not affected, because the prediction already limits battery draw to inverter_limit (prediction.py:1141). If the requested value should be right at the source too, one option is to bind inverter_limit_charge/inverter_limit_discharge to inverter_size with set_arg_auto(..., overwrite=False). Capping inside number_event_handler would instead make write_and_poll_value fail its read-back and set had_errors every cycle.
• Not done: stepping down after a refusal. The issue asks to step down to the capped value when a write is refused. Now that the first write already is the capped value, that is covered. There is no further step-down if the capped value is also refused.
• Not done: leaving a 0 A discharge slot disabled. An earlier revision stopped enabling a discharge slot while its current was refused and read back 0 A. It was dropped at review to keep the change to the cap alone and leave the freeze-charge and hold paths untouched. Freeze charge sets discharge_rate = 0 (execute.py:517), and the discharge slot can be armed ahead of an export window while that is still in force. If the capped current is refused, a V2 slot can still sit at 0 A.
• Unconfirmed: 75 A on the affected unit. Whether 75 A is accepted there is not yet proven. Same-model sites accept 70 A and 62.5 A. A 16S pack converts at 51.2 V, which gives 70 A. A 48 V (15S or unknown) pack gives 75 A. solis_nominal_voltage sets the conversion voltage directly.
• HV packs. These convert at live voltage, so the rating moves with state of charge. Flooring to whole amps limits the rewrites to crossings of an amp boundary.
• New log line. V2 inverters whose battery limit exceeds their rating log Capping slot currents on <sn> at <n>A once per control cycle.
• One-off V1 rewrite. A V1 install whose decoded slot currents exceed the rating gets one CID 103 rewrite after upgrading.
• Recovery SoC edge cases. A reading above 100 with no known over-discharge SoC is now treated as unknown, and the target is left alone. Before, the target was clamped to the bogus value. A missing reading behaves as before.
• Blast radius. GitNexus impact: resolve_discharge_soc_floor LOW (one caller), encode_time_windows LOW (one caller, the V1 branch), write_time_windows_if_changed LOW. The index lists no callers for it; by search its only production caller is run(). detect_changes: LOW, no affected processes.
• No triage comment. The issue has no bot triage comment. This PR works from the root cause the issue itself gives, checked against main at 8b9fe02.

🤖 Generated with Claude Code

[Copilot]

Copilot review overview

🟡 Changes recommended

Moderate issues remain in V1 battery-limit handling and conservative current-rating truncation.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Fix grammatical typo in current-limit comment · New

What changed in this PR

Adds Solis inverter-rated current caps and fallback handling for implausible recovery SoC values.

Changes:

• Caps V1/V2 slot currents to inverter ratings.
• Validates recovery SoC readings with a safe fallback.
• Adds focused regression tests.

File	Summary
apps/​predbat/​tests/​test_solis.py	Adds coverage for current caps and recovery SoC handling.
apps/​predbat/​solis.py	Implements current limiting and recovery SoC validation.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.