Cooling Tower Test Matrix

Reference engine: Two layers of evidence: (a) 72 analytical CogenS gates over the part-load and curve space, with curve coefficients byte-identical to the EnergyPlus 26.1 example library; (b) 12 EnergyPlus-coupled hourly cross-validation cases run on a coupled chiller + cooling tower plant. ASHRAE Guideline 14-2023 scoring on layer (b).

Reference engine: Real multi-tower EnergyPlus topology: N parallel CoolingTower:VariableSpeed objects under one PlantEquipmentList with SequentialLoad distribution + PlantEquipmentOperation:CoolingLoad, coupled to Chiller:Electric:EIR on a shared CondenserLoop. Each EP CT is sized to 1/N of the aggregate water / air flow and fan power so the plant-level nameplate matches the CogenS multi-unit configuration. Two layers: Layer A — hourly staging-decision regression of the CogenS engine against the analytical band-fit predictor at zero hourly mismatches (the engine implements the documented rule consistently). Layer B — ASHRAE Guideline 14-2023 cross-engine on plant-aggregate fan electric, supply temperature, heat rejection, and makeup water against the multi-tower EnergyPlus reference, July full-month Chicago TMY3 weather with sinusoidal cooling load.

Reference engine: Two layers. Layer A — 24 analytical cases gating the hourly free-cooling mode classifier (Mode 0 when supply > 10 C, Mode 1 full FC when cooling load <= qecomax and supply <= 10 C, Mode 2 partial FC when cooling load > qecomax and supply <= 10 C) at zero hourly mismatches, plus a gate that the engine applies ONE consistent free-cooling effectiveness across every full-FC hour. No reference engine. Layer B — 1 EnergyPlus cross-engine case at Chicago January TMY3 (744 hourly steps) against a plant that now carries a HeatExchanger:FluidToFluid waterside economizer, listed ahead of the chiller with CoolingSetpointOnOffWithComponentOverride locking the chiller out while the exchanger carries the load. Before that object was added the reference had NO economizer at all — its chiller ran 744 of 744 hours while CogenS free-cooled, so the two engines were modelling different plants and the case was not scoreable. It is now, and it FAILS: see the gates below.

Reference engine: Two-layer evidence pack. Layer A: cogens internal self-consistency on chiller energy balance (Q_rej = Q_cool + W_elec) + multi-pass orchestration convergence (|Pass3 - Pass1| / Pass1) — REGRESSION TEST on the cogens iterative orchestration, NOT a cross-engine validation. Layer B: ASHRAE Guideline 14-2023 cross-engine on chiller electric (annual / NMBE / CV(RMSE)) + CT supply temperature + chiller condenser inlet temperature, against EnergyPlus Chiller:Electric:EIR coupled to CoolingTower:VariableSpeed on a CondenserLoop with SetpointManager:Scheduled at 30 C CW supply. Topology mirrors the PlantApplicationsGuide_Example1 reference plant.

Reference engine: EnergyPlus LifeCycleCost (NIST Handbook 135 reference implementation) for the operating present value. Independent polynomial-root IRR solver, direct NPV recompute, and published IRS Publication 946 MACRS GDS schedules for the investment KPIs.

Reference engine: EnergyPlus 26.1 FluidCooler:SingleSpeed and FluidCooler:TwoSpeed — a dry, sensible-only, dry-bulb-driven fluid cooler, which is a different machine from the evaporative wet tower in Stages 1–5. Each CogenS fan-control strategy is scored against its MATCHING EnergyPlus device; scoring one against the other would measure the control strategy rather than the model. Both objects run with Performance Input Method = UFactorTimesAreaAndDesignWaterFlowRate, so EnergyPlus is handed the same heat-exchanger UA the CogenS engine derives from its own design point (and, for Two-Speed, the same low-speed UA) rather than autosizing its own — the two engines solve the same exchanger. Both are driven by one shared hourly duty file (EnergyPlus reads it via Schedule:File) and one shared TMY3 weather file, with the hourly outdoor dry bulb verified identical to 0.00000 °C.