SAFETY-119: no "not keeping up" when the loop takes the wire back

When the peak guard or a maintenance charge hands the wire back, the loop's
integrator is re-seeded from that owner's command and kicked by a meter that
does not show it yet. It crossed its rail for a cycle or two and raised
goodwe_integrator_bound while the battery gave what it was told.

The loop's integrator bound is not counted until the plant's settle
(SHORT_CYCLES, SHORT_SPAN_S) has passed since the hand-back. An over-rating
always counts; a bound that lasts warns after the settle. No command
changes and no text changes. test_safety119.py; DECISIONS entry.

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
glenn schrooyen
2026-10-19 14:14:08 +02:00
co-authored by Claude
parent 75f0804f9d
commit c4e15ec4cd
3 changed files with 404 additions and 1 deletions
@@ -0,0 +1,27 @@
# SAFETY-119 - no "not keeping up" when the loop takes the wire back
**Status: for review.** Tests: `test_safety119.py`.
**Cause.** When another owner (the peak guard, a maintenance charge) has set
the wire and the loop wins again, `cycle()` re-seeds the integrator from that
owner's command, and `compute()` adds `gain * error` from a meter that does
not show that command yet. On the TEST-23 rig (SAFETY-118 I-2) the guard's
last-seconds step went from -2000 W to +429 W. The next cycle's ask crossed
the +2500 W rail and beat the guard's floor, so SAFETY-85 rule 1 counted it
as the loop's own bound and raised `goodwe_integrator_bound`. The battery gave
what it was told; the cap held.
**Decision.** On the cycles after a hand-back, the loop's integrator bound is
not counted until the plant's settle has passed (`SHORT_CYCLES` cycles and
`SHORT_SPAN_S`, the same window SAFETY-85 grades a battery on). After that it
counts as before. An over-rating (`clamped`, a bug) always counts. No command
changes, and no text was added or changed.
**Kept.** The loop in charge at its rail warns on the same cycle as before. A
loop still at its rail after the window warns then (about 22 s later): delay,
not loss. The start and a restart after a stop are not hand-backs.
**Not changed.** The kick itself is control behaviour: for a few seconds at
a quarter's edge the loop asks up to its rail. In `test_safety119.py` part 1
that is more discharge than the guard asked at a quarter's end, and about
1700 Ws more charge (2 W on that quarter's average) at the next one's start.
+30 -1
View File
@@ -1272,6 +1272,10 @@ class Controller:
self._loop_decision = None
self._asked_hist: dict = {}
self._short_run = 0
# SAFETY-119: did another owner set the wire last cycle, and since
# when (monotonic s, owned cycles) has the loop had it back.
self._wire_lost = False
self._handback = None
self.stopping = False
# SAFETY-61. time.monotonic() at the end of the last run_control() pass.
self.loop_pass_mono = None
@@ -5813,6 +5817,7 @@ class Controller:
if d is None:
self._short_run = 0
self._asked_hist = {}
self._wire_lost, self._handback = False, None
return
unmet = self.allocation.unmet_w if self.allocation is not None else 0
cut = (resolution.bound_by or "").split("(")[0]
@@ -5838,7 +5843,9 @@ class Controller:
and delivery_short(asked[inv.name], inv.batt,
self.tuning.saturation_w)]
self._short_run = self._short_run + 1 if short else 0
if owns and self.loop_clamped[0]:
settling = self._settling(owns, now)
if owns and self.loop_clamped[0] and not (
settling and not self.loop_clamped[2]):
return
engaged = self._short_run >= self.tuning.saturation_cycles
named = len(self.inverters) > 1
@@ -5847,6 +5854,28 @@ class Controller:
if engaged else "", False)
self.loop_clamp_facts = ()
def _settling(self, owns, now) -> bool:
"""SAFETY-119: is the loop within the plant's settle (SHORT_CYCLES
cycles and SHORT_SPAN_S) of taking the wire back from another owner?
Its integrator is re-seeded from that owner's command and kicked by a
meter that does not show it yet, so its bound there is not counted.
DECISIONS/SAFETY-119-bound-after-hand-back.md."""
if not owns:
self._wire_lost, self._handback = True, None
return False
if self._wire_lost:
self._wire_lost, self._handback = False, (now, 0)
if self._handback is None:
return False
t0, n = self._handback
n += 1
self._handback = (t0, n)
if n >= SHORT_CYCLES and (now is None or t0 is None
or now - t0 >= SHORT_SPAN_S):
self._handback = None
return False
return True
async def check_output_clamp(self) -> None:
"""SAFETY-03. Alarm whenever a clamp actually engaged this cycle.
+347
View File
@@ -0,0 +1,347 @@
"""Runnable check for SAFETY-119: no "Battery is not keeping up" when the
loop takes the wire back from another owner. `python3 test_safety119.py`
During a maintenance charge with no usable peak forecast, the measured peak
guard steps in during a quarter's last seconds and hands back. On the cycle
the loop gets the wire back, its integrator is re-seeded from the other
owner's command and kicked by a meter reading taken before the battery moved.
It crossed its rail for one or two cycles, the loop's ask beat the one-sided
limit, so the cycle counted as the loop's own and raised
`goodwe_integrator_bound`. The battery gave what it was told. Now the loop's
bound does not count until the plant's settle (SHORT_CYCLES cycles and
SHORT_SPAN_S) has passed since the hand-back.
DECISIONS/SAFETY-119-bound-after-hand-back.md.
1. The rig's shape (1 s meter that lags the battery, 1 s battery lag, house
800 W, rating 2500 W, limit 2500 W, margin 50, charge 2000 W, no
forecast), 2 h, three seeds: no notification, the cap held, every command
the same as without the fix; without the fix it warns.
2. A genuine integrator bound still warns: the loop in charge with the house
over the rating (at once, as before), and the loop at its rail after a
hand-back (after the settle, not lost).
3. With a usable forecast: the same commands and the same alarms as without
the fix.
4. The rule at its boundaries (`Controller._settling`).
Mutations (MUTATION.md, purged copies) are listed in the pull request.
"""
import logging
import math
import os
import random
import sys
import time
from datetime import timedelta
HERE = os.path.dirname(os.path.abspath(__file__))
sys.path.insert(0, HERE)
sys.path.insert(0, os.path.join(HERE, "tools"))
logging.disable(logging.WARNING)
import opt11_identity as rig # noqa: E402
import safety79_identity as s79 # noqa: E402
import safety84_identity as s84 # noqa: E402
import app.main as app_main # noqa: E402
from app import messages # noqa: E402
from app.main import Controller # noqa: E402
from app.p1 import make_sample, split_signed # noqa: E402
fails = []
OPT0 = dict(s84.OPT)
CAP = 2500.0
QUARTER = 900.0
def check(name, cond):
print(f" {'ok ' if cond else 'FAIL'} {name}")
if not cond:
fails.append(name)
async def _call(self, *a):
"""The rig's Home Assistant, recording each notification (time, id)."""
if len(a) >= 3 and a[1] == "create" and isinstance(a[2], dict):
self.__dict__.setdefault("notes", []).append(
(self.t, a[2].get("notification_id")))
return True
def _power(self, u, lag=1.0):
"""The TEST-23 plant's battery: 800 W/s ramp, 1 s first-order lag."""
st = self.__dict__.setdefault("_plant", {})
a, tgt, tl = st.get(u, (0.0, 0.0, self.t))
for _ in range(int(round((self.t - tl) / 0.1))):
d = self.cmd[u] - tgt
tgt += math.copysign(min(abs(d), 80.0), d)
a += (tgt - a) * (1 - math.exp(-0.1 / lag))
st[u] = (a, tgt, self.t)
if self.cut(u, self.t) and self.cmd[u] >= -4.0:
return -4.0
return float(round(a))
_real_number = rig._Hass.number
async def _number(self, e, invert=False):
"""sensor.fc: a usable demand forecast of 800 W (part 3)."""
if e == "sensor.fc":
return 800.0
return await _real_number(self, e, invert)
rig._Hass.call, rig._Hass.power, rig._Hass.number = _call, _power, _number
rig.TICK = 1.0
DELAY = [1]
def _feed(c, t):
"""The P1 meter `DELAY` ticks behind the battery, as on the rig (the
meter reached HA after the battery reading). Records the true grid."""
h = c.hass.__dict__.setdefault("grid_true", [])
h.append((t, c.hass.grid()))
g = h[max(0, len(h) - 1 - DELAY[0])][1]
imp, exp = split_signed(g)
c.p1.submit(make_sample("ha_signed", imp, exp, phases=1,
ingest_ts=rig.BASE + timedelta(seconds=t),
ingest_mono=time.monotonic()))
s79._feed = _feed
def house(seed, base=800.0, step=None):
"""800 W +- 30 W, a new value each second; `step` = (t, W) from then."""
rnd = random.Random(seed)
noise = [rnd.uniform(-30.0, 30.0) for _ in range(4000)]
def load(t):
if step is not None and t >= step[0]:
return step[1]
return base + noise[int(t) % 4000]
return load
def run(load, *, dur=7200.0, delay=1, extra=None, cheap=True, cut=None,
charging=True):
"""One N=1 run; `charging=False`: no maintenance, the loop in charge."""
extra = dict(extra or {})
s84.OPT = dict(OPT0, peak_cap_w=int(CAP), peak_margin_w=50,
maintenance_charge_w=2000)
if not charging:
extra.update(maintenance_enabled=False, peak_cap_w=20000)
for k in [k for k in extra if k in s84.OPT]:
s84.OPT[k] = extra.pop(k)
DELAY[0] = delay
try:
return s84.run(1, load=load, cheap=cheap,
store=(s79.store_charging(1) if charging
else rig._Store()),
soc=lambda u, t: 60.0, dur=dur, cut=cut,
extra=dict({"max_w": 2500, "slew_w": 0}, **extra))
finally:
s84.OPT = OPT0
def patched(obj, name, value):
def wrap(fn):
real = getattr(obj, name)
setattr(obj, name, value)
try:
return fn()
finally:
setattr(obj, name, real)
return wrap
no_fix = patched(Controller, "_settling", lambda self, owns, now: False)
def notes(c):
return [t for t, i in getattr(c.hass, "notes", [])
if i == messages.ID_LOOP_BOUND]
def quarters(c):
"""Each whole quarter's true average import (W)."""
q = {}
for t, g in c.hass.grid_true:
q.setdefault(int(t // QUARTER), []).append(g)
return [sum(v) / len(v) for v in q.values() if len(v) == QUARTER]
def commands(rows):
return [(r[0], r[1], r[2]) for r in rows]
def guard_steps(c):
return sum(1 for e in c.events if "PEAK_GUARD:" in e)
# ============================================================================ #
print("1. a maintenance charge with no forecast, the guard steps in at each "
"quarter's end, 2 h")
NEW, OLD = {}, {}
for seed, delay in ((1, 1), (2, 1), (3, 2)):
NEW[seed] = run(house(seed), delay=delay)
OLD[seed] = no_fix(lambda: run(house(seed), delay=delay))
print(f" seed {seed} meter {delay} s behind: notifications "
f"{notes(NEW[seed][1])} (without the fix {notes(OLD[seed][1])}), "
f"highest quarter {max(quarters(NEW[seed][1])):.0f} W, guard "
f"step-ins in the last 40 log rows {guard_steps(NEW[seed][1])}")
check("⭐ no 'Battery is not keeping up' notification, cycle or episode",
all(notes(c) == [] and c.clamp_wd.bound == 0
and c.clamp_wd.bound_episodes == 0 for _r, c in NEW.values()))
check("⭐ the peak cap is held: every quarter's average is under 2500 W",
all(len(quarters(c)) == 8 and max(quarters(c)) < CAP
for _r, c in NEW.values()))
check("⭐ control is unchanged: every command the same as without the fix",
all(commands(NEW[k][0]) == commands(OLD[k][0]) for k in NEW))
check("...the charge ran and the guard stepped in near each quarter's end",
all(any(r[1] == -2000.0 for r in rows) and any(
r[-1][1] and r[-1][1][0] and r[0] % QUARTER >= QUARTER - 40
for r in rows) for rows, _c in NEW.values()))
check("non-vacuous: without the fix every run warns, at a quarter's edge",
all(len(notes(c)) >= 1 for _r, c in OLD.values())
and all(min(t % QUARTER, QUARTER - t % QUARTER) <= 40
for _r, c in OLD.values() for t in notes(c)))
# ============================================================================ #
print("\n2. a genuine integrator bound still warns")
_rail = run(house(1, base=4000.0), dur=600.0,
charging=False)
_rail_old = no_fix(lambda: run(house(1, base=4000.0), dur=600.0,
charging=False))
print(f" loop in charge, house 4000 W over the 2500 W rating: "
f"{notes(_rail[1])}, without the fix {notes(_rail_old[1])}")
check("⭐ the loop in charge at its rail warns (log and notification), at "
"the same cycle as before",
len(notes(_rail[1])) >= 1 and notes(_rail[1]) == notes(_rail_old[1])
and _rail[1].clamp_wd.bound == _rail_old[1].clamp_wd.bound
and any("OUTPUT_CLAMP episode" in e for e in _rail[1].events))
# The charge sets the wire until 600 s, then a solar surplus of 6000 W: the
# loop takes the wire back and stands at its -2500 W rail.
_hb = run(house(1, step=(600.0, -6000.0)), dur=1200.0,
extra={"peak_cap_w": 20000})
_hb_old = no_fix(lambda: run(house(1, step=(600.0, -6000.0)), dur=1200.0,
extra={"peak_cap_w": 20000}))
_first = notes(_hb[1])[:1]
print(f" hand-back to a loop at its rail at 600 s: first notification "
f"{_first}, without the fix {notes(_hb_old[1])[:1]}; command at the "
f"end {_hb[0][-1][1]}")
check("⭐ the loop at its rail after a hand-back still warns, once the "
"settle has passed (12 cycles, 22 s): delay, not loss",
_first and 600.0 + app_main.SHORT_SPAN_S <= _first[0] <= 640.0
and _hb[0][-1][1] == -2500.0
and _hb[1].clamp_wd.bound > 100)
check("...without the fix it warned at once (non-vacuous)",
notes(_hb_old[1])[:1] and notes(_hb_old[1])[0] < 600.0
+ app_main.SHORT_SPAN_S)
# The same charge with no forecast; at 1795 s, under the guard's step-in at
# the quarter's end, the house goes to 4000 W, over the 2500 W rating.
_up = run(house(1, step=(1795.0, 4000.0)), dur=1900.0)
_up_old = no_fix(lambda: run(house(1, step=(1795.0, 4000.0)), dur=1900.0))
print(f" guard step-in, then the house over the rating from 1795 s: "
f"{notes(_up[1])}, without the fix {notes(_up_old[1])}")
check("⭐ a genuine bound in the rig's own scenario still warns, within the "
"settle of the guard's hand-back",
notes(_up[1]) and 1795.0 <= notes(_up[1])[0] <= 1795.0 + 40.0
and any("OUTPUT_CLAMP episode" in e for e in _up[1].events))
# ============================================================================ #
print("\n3. with a usable forecast (800 W): unchanged")
FC = {"peak_forecast_entity": "sensor.fc"}
_fc = run(house(1), extra=FC)
_fc_old = no_fix(lambda: run(house(1), extra=FC))
print(f" notifications {notes(_fc[1])} (without the fix "
f"{notes(_fc_old[1])}), highest quarter "
f"{max(quarters(_fc[1])):.0f} W, charge "
f"{min(r[1] for r in _fc[0][:300]):.0f} W")
check("⭐ the same commands and the same alarms as without the fix, and "
"the cap held",
commands(_fc[0]) == commands(_fc_old[0])
and notes(_fc[1]) == notes(_fc_old[1])
and _fc[1].clamp_wd.engagements == _fc_old[1].clamp_wd.engagements
and max(quarters(_fc[1])) < CAP)
check("...the forecast was used (the charge is shaped under -2000 W)",
-2000.0 < min(r[1] for r in _fc[0][:300]) < 0.0)
# ============================================================================ #
print("\n4. the rule at its boundaries")
class _C:
_settling = Controller._settling
def __init__(self):
self._wire_lost, self._handback = False, None
def seq(c, steps):
return [c._settling(o, t) for o, t in steps]
N, SPAN = app_main.SHORT_CYCLES, app_main.SHORT_SPAN_S
_c = _C()
check("from the start, owned every cycle: never settling (no hand-back)",
seq(_c, [(True, float(i)) for i in range(30)]) == [False] * 30)
_c = _C()
_out = seq(_c, [(False, 0.0)] + [(True, 2.0 * i) for i in range(1, N + 3)])
check(f"after another owner: settling for {N - 1} owned cycles, then not "
f"(12 cycles at 2 s span {SPAN:.0f} s)",
_out == [False] + [True] * (N - 1) + [False] * 3)
_c = _C()
_out = seq(_c, [(False, 0.0)] + [(True, 0.5 * i) for i in range(1, 60)])
_end = _out.index(False, 1)
check(f"at a 0.5 s meter the span decides: settling until {SPAN:.0f} s",
_end == 1 + int(SPAN / 0.5) and all(_out[1:_end]))
_c = _C()
_out = seq(_c, [(False, 0.0), (True, 2.0), (True, 4.0), (False, 6.0),
(True, 8.0), (True, 10.0)])
check("losing the wire again restarts the window",
_out == [False, True, True, False, True, True]
and _c._handback == (8.0, 2))
_c = _C()
check("no monotonic clock: the cycle count alone decides",
seq(_c, [(False, None)] + [(True, None)] * (N + 1))
== [False] + [True] * (N - 1) + [False] * 2)
# An over-rating (a command over `max_w`) is a bug, never settled away.
_ctl = _hb[1]
_ctl.loop_clamped = (True, "x", True)
_ctl._wire_lost, _ctl._handback = True, None
class _Res:
winner, bound_by, target_w = "loop", "", 0.0
unit_bounds = {}
class _D:
held, target_w = False, 0.0
_ctl._loop_decision, _ctl.allocation = _D(), None
_ctl._loop_alarm_input(_Res())
check("an over-rating on a hand-back cycle still counts (bug=True kept)",
_ctl.loop_clamped == (True, "x", True) and _ctl._handback is not None)
_ctl.loop_clamped = (True, "y", False)
_ctl._loop_alarm_input(_Res())
check("...while the integrator bound on the next cycle is not counted",
_ctl.loop_clamped[0] is False)
_ctl._wire_lost, _ctl._handback = True, (1.0, 3)
_ctl._loop_decision = None # stopped or blind
_ctl._loop_alarm_input(_Res())
check("a stop clears it: the next start is not a hand-back",
(_ctl._wire_lost, _ctl._handback) == (False, None))
print()
if fails:
print(f"FAILED: {len(fails)} check(s)")
for f in fails:
print(f" - {f}")
sys.exit(1)
print("all SAFETY-119 checks passed")