Chilled-water plant reset
The hero sequence. It relaxes the chilled-water plant while the air handlers are satisfied and
tightens it again when they ask for more cooling. One reset value r between 0 and 1 drives two
setpoints: at r = 1 the plant runs at design; as r falls the supply water warms first, and
only then does the pressure setpoint drop.
Written in our own words from the behavior of the Modelica Buildings Library’s plant reset sequence. Nothing is copied from it.
Interface
Section titled “Interface”| Point | Type | Meaning |
|---|---|---|
nReqRes |
integer input | Number of chilled-water reset requests from the air handlers |
u1Ena |
boolean input | The plant is enabled |
u1StaPro |
boolean input | A chiller staging change is in progress |
TSupSet |
real output, K | Chilled-water supply temperature setpoint |
dpSet |
real output, Pa | Chilled-water differential pressure setpoint |
Behavior in brief
Section titled “Behavior in brief”- Disabled:
r= 1, both setpoints at design (278.71 K, 239,125 Pa, BOPTEST’s baseline values), at once. - Start delay: after an enable,
rholds at 1 for 900 s plus one 300 s update period. - Updates on a 300 s grid: trim by −0.02, or, when requests exceed 2, respond by +0.03 per
extra request (capped at +0.07).
rstays within [0, 1]. - Staging hold: a staging change freezes updates for at least 900 s, longer if staging lasts.
- Mapping: supply temperature runs from 282.15 K at
r≤ 0.5 to design atr= 1; pressure runs from 95,650 Pa atr= 0 to design atr≥ 0.5.
The task also states the edge instants (when the first change lands, which update a staging start freezes), because the prose alone leaves them open.
Validation
Section titled “Validation”- 10 scenarios from the independent reference model: start delay and trim cadence, ignore/respond/cap, saturation and release, short and long staging holds, a second staging start, changes between updates, disable and re-enable, recovery from the floor.
- A 33-block hand-built graph passes all 10; 12 deliberate mistakes (such as a hold whose false duration keeps its default, or uncapped respond) each fail at least one.
How it did
Section titled “How it did”- Muse Spark: 3 of 3 live sessions verified with loop-aware repair and the loop rule; harder without vector feedback. See Benchmark results.
- On BOPTEST with air handler requests, 10 hours: HVAC energy −4.7 %, chiller energy −14.2 %, thermal discomfort +2.0 %. See Runtime results.
Full spec: specs/chw_plant_reset/spec.md