DHW Heat Pump Water Heater Module Validation

Stage 1 — Steady-state single-tank shakedown

Single heat-pump water heater at constant ambient and constant draw over 30 days (720 h). Two condenser archetypes (wrapped-condenser residential, pumped-condenser stratified) × three ambient temperatures (10 / 20 / 30 °C) × two continuous draw rates, each sized inside the heat pump's sustainable capacity so the resistance backup never fires. EnergyPlus defines the thermal load; the two engines are then compared on heat-pump compressor electricity. Because a 30-day constant-draw run has neither an annual nor a monthly dimension, only the total-energy gate applies — this stage is a model shakedown, not the headline validation.

What this stage is for

Stage 1 holds ambient and draw constant so that any disagreement is attributable to the equipment model rather than to load or weather. It is deliberately synthetic. Its value was diagnostic: it surfaced and closed four defects — a draw schedule sized beyond heat-pump capacity, a dry-bulb temperature fed to a wet-bulb performance curve, a humidity ratio misread as relative humidity, and un-normalised curve coefficients — before any annual result was trusted.

Why the pumped-condenser cases fail here and pass at annual scale

All six pumped failures are under-predictions of 6.74% to 11.10% on a 0.189 m³ (50 gal) tank. The pumped condenser draws from the tank's bottom node; at that small volume the sump node runs cooler than EnergyPlus's, raising the modelled coefficient of performance and lowering predicted electricity. The same archetype on the 3.0 m³ commercial bank in Stage 2 does not exhibit it. The limitation is therefore specific to small-volume pumped-condenser configurations.

Stage 2 — Annual realistic building load, four climates

Full 8,760-hour annual simulation driven by the DHW draw of the DOE / ASHRAE 90.1-2019 Large Hotel prototype building, run as shipped so the load is an authority's rather than an assumption. Two condenser archetypes × four climates (Chicago 5A, San Francisco 3C, Tampa 2A, Golden CO semi-arid) × two evaporator ambient sources (indoor mechanical room, outdoor weather-driven). Equipment is a commercial 4 × 20 kW bank with 3.0 m³ storage, sized from the measured load against the element-trip threshold rather than a rule of thumb. This is the headline validation.

How the load was established

The DOE Large Hotel prototype is run unmodified and its own water heater's hourly use-side heat delivery becomes the DHW load. Nothing about the draw profile is authored for this study. The resulting load is 161 to 219 MWh per year with a load factor near 0.43 — roughly 55 litres per room per day across about 200 rooms. Its variation across the four climates is pure mains-water temperature: predicted ratios of 1.000 / 0.915 / 0.734 against observed 1.000 / 0.918 / 0.739, which independently confirms that climate enters only where it physically should.

Why the model is scored on annual and monthly energy

A techno-economic analysis rests on annual energy and monthly utility bills, not on which hour within a month a compressor happens to fire. Hourly agreement was measured and found to be dominated by intra-hour phase differences between two tank solvers rather than by any energy disagreement: hourly CV(RMSE) of 43% to 104% collapses to 1.8% to 9.5% when the identical arrays are binned monthly, while NMBE passes 16 of 16 at both resolutions. Hourly and sub-hourly agreement are therefore outside the scope of this validation, and are stated as such rather than quietly omitted.

Timestep-sensitivity control test — all 16 scenarios

For a single-speed compressor the part-load ratio is its fractional on-time, and the part-load efficiency curve penalises low part-load. That penalty is evaluated at the simulation timestep, so a coarser timestep averages the on-time into a lower part-load ratio and charges a larger penalty. EnergyPlus runs 15-minute steps; the model runs 60-minute steps by default. This yields a prediction that can be falsified: the wrapped-condenser archetype carries a FLAT part-load curve and therefore cannot be timestep-sensitive, while the pumped-condenser archetype carries a sloped one and must shift. Every scenario was re-run at 15 minutes to test it.

The result is unambiguous. Wrapped: mean +0.19%, range +0.10% to +0.28%, across all eight scenarios — no timestep sensitivity, as the flat part-load curve requires. Pumped: mean −6.21%, range −8.13% to −3.90%, negative in all eight. The two families do not overlap at any point, and there is no counter-example.

One consequence is reported because it bears on how the result should be read: the finer timestep does not produce a clean sweep. At 15 minutes the score is also 15 of 16, but a DIFFERENT scenario falls outside the band (pumped · Golden · outdoor, at −5.13%). Moving to the finer timestep is therefore a re-calibration of the part-load treatment rather than a correction, which is why it has not been adopted as a fix. The practical figure for a user is that pumped-condenser annual energy carries a timestep sensitivity of roughly 6% to 8%, and wrapped-condenser carries none.