Heat-Pump Boiler Module Validation

Stage 1 — Core heat-pump model, air-source and water-source

A single heat-pump boiler on a closed hydronic loop across the full operating range: two source types (outdoor air and a water / waste-heat source), four leaving-water setpoints from 35 to 65 degrees Celsius, eight source temperatures from minus 20 to plus 25 degrees Celsius, four part-load ratios, three compressor control types (variable speed, staged, single speed), and two machine sizes. This is the stage that proves the capacity and efficiency curves themselves — everything above it inherits this result.

A heat-pump boiler is not a boiler with a different fuel. Its capacity and its efficiency both move with two temperatures at once — how cold it is outside, and how hot the water has to leave — and they move in opposite directions to what a project needs. The coldest hour of the year is when the building wants the most heat and when the machine can make the least of it. Stage 1 establishes that the CogenS engine reproduces that behaviour, point for point, against the same equipment modelled in EnergyPlus.

What the matrix covers

336 scenarios sweep two source types, four leaving-water setpoints (35, 45, 55 and 65 degrees Celsius), eight source temperatures from minus 20 to plus 25 degrees Celsius, four part-load ratios, three compressor control types and two machine sizes (80 kW and 400 kW). Air-source machines see outdoor air as the source; water-source machines see a fixed-temperature water or waste-heat stream. Both engines read the identical curve coefficients, so any difference is a difference in how the two models apply them.

Headline result

Every one of the 336 scenarios passes on all eight gates. Total heat delivered lands between +0.01 to +0.12 percent; total electricity between +0.01 to +0.12 percent; hourly NMBE never exceeds 0.13 percent against a 10 percent band. In practical terms the two engines are reporting the same machine to within a fifth of one percent, which is far below the uncertainty in the manufacturer data that feeds either of them.

What we improved in CogenS while running this validation

Two findings came out of this stage and both are now in the production engine. First, the temperature axes of the capacity and efficiency curves run the opposite way to the domestic hot-water heat-pump convention: for a heat-pump boiler they are indexed on the temperature the water LEAVES at and the temperature the source enters at. Getting that backwards produces a model that looks plausible and is wrong at every ambient. Second, the published curves carry output limits that must be applied after evaluation; without them an extrapolated curve can return a coefficient of performance below one, which is thermodynamically impossible for a heat pump. Both are now enforced, and the engine additionally refuses to report a coefficient of performance below one under any input.

Stage 1B — Full-year 8,760-hour run on real weather

The whole plant run for a complete year against real TMY3 weather in four climates, with the hourly heating load supplied to both engines as one identical file so a load difference can never be mistaken for a model difference. Chicago and Golden are cold; San Francisco is mild marine; Tampa is the low-load extreme at 17.9 MWh of annual heating. The plant is sized to each climate's peak, as a real project would be. No climate is excluded.

Every other stage is a controlled probe at one condition. This is the one where the mechanisms interact the way they do on a real project: capacity and efficiency fall together as the weather turns, the operating envelope caps how hot the water can get on the coldest hours, the bank stages up and down as the load swings, and the auxiliary picks up whichever limit bites first — 8,760 times, in sequence. It is also the stage whose numbers a customer actually cares about: annual electricity, annual gas, and how much of the heating load the heat pump carried across a real year.

Four climates, spanning a factor of twelve

Headline result

All 8 annual runs pass every gate. Annual heat-pump electricity lands between -0.85 to +0.47 percent of the EnergyPlus result, and the two engines agree on how much of the year's heating the heat pump carried — as opposed to the gas auxiliary — to within 0.08 percent. That split is the number a heat-pump boiler project turns on, because it decides the electricity bill, the gas bill and the emissions result simultaneously.

What we improved in CogenS while running this validation

This stage found a real modelling gap and it is now fixed. On a bank of modules sharing a common header at constant flow, a module that is running heats only its own branch while the idle modules pass cooler water that mixes back in downstream. The running module therefore has to send its water out hotter than the loop's own supply temperature, which costs efficiency. The engine had been evaluating the performance curves at the loop supply temperature, which is correct only when the idle modules are valved off. EnergyPlus reports one module of two, running at 21.7 kW into a 41.4 degree return, leaving at 48.5 degrees against a 45 degree loop setpoint; the corrected engine predicts 48.5 degrees. Left uncorrected it overstated seasonal efficiency by 5 to 6 percent. The hydronic arrangement is now an explicit input, because it is a real design choice and it is worth that much on the operating cost.

The correction also explains why this only surfaced at the annual stage. A single-condition probe holds the water temperature fixed by construction, so the question never arises; it took a full year on a properly sized hydronic loop to expose it.

Stage 2 — Multi-module bank staging

Banks of two, three and four identical modules under four load levels from 20 to 95 percent of bank capacity, at four source temperatures, with two compressor control types. This proves the engine stages the right number of modules and charges the right electricity for the way they share the load.

Commercial heat-pump boilers are almost always specified as a bank of modules rather than one large machine, because modularity is how the plant holds efficiency at part load. Stage 2 checks that the engine brings on the right number of modules and prices the result correctly.

Headline result

All 96 scenarios pass. Delivered heat matches to +0.04 to +0.19 percent — the two engines stage identically. Bank electricity lands between -4.20 to +9.71 percent, inside the 15 percent federal band, for the reason set out next.

Where the two engines model the same equipment differently

The gap is a load-sharing convention, not an error in either engine. When a bank runs partly loaded there are two defensible control laws: share the load equally across every running module, or fill each module before starting the next. Equal sharing keeps each module at a lower part-load ratio; sequential filling runs some at full output and one at the remainder. Both are used in real control systems, both deliver identical heat, and they differ on electricity by up to 11 percent because the cycling penalty is not linear in load. CogenS supports both and names which one a project is using. Because this is a documented difference between two correct models rather than a modelling error, bank electricity is scored under the federal FEMP Measurement and Verification band, which ASHRAE Guideline 14-2023 provides for exactly this case. Delivered heat — the quantity a building actually experiences — is scored strictly, and matches to under two tenths of one percent.

Stage 3 — Operating envelope against published vendor data

The maximum water temperature a machine can actually produce is not a fixed nameplate number — it falls as the source gets colder, and that is what decides how much auxiliary heat a project needs on its design day. This stage scores the engine's envelope model directly against 18 operating points published by eight manufacturers across air-source, water-source, waste-heat and carbon-dioxide machines.

This is the behaviour that separates a heat-pump boiler from every other heating plant, and the one a simplified model gets wrong. A Trane air-source machine makes 140 degrees Fahrenheit of fluid at 55 degrees Fahrenheit outdoor air, but only 90 degrees Fahrenheit at 0 degrees Fahrenheit. A model that assumes the machine always makes its rated water temperature will show a design day the plant cannot actually deliver, and will under-size the auxiliary that has to cover the gap. The economics of the project turn on exactly that gap.

Why this stage is scored against vendor data rather than EnergyPlus

EnergyPlus provides a maximum-supply-temperature curve field on its heat-pump object, and the validation work established across eight separate plant configurations that the field does not restrict the machine's output in the released version — the machine delivers its curve capacity regardless of what the envelope curve is set to. Scoring the envelope against EnergyPlus would therefore measure nothing. The manufacturers' own published envelopes are the stronger reference here, and they are public: any engineer can open the same catalog page and check the number.

Headline result

All 18 published points reproduce to 0.00 degrees Celsius against a 1.0 degree band, spanning air-source machines from minus 20 to plus 55 degrees Celsius source temperature, transcritical carbon-dioxide machines holding 90 degrees Celsius supply down to minus 25 degrees Celsius, and a waste-heat machine holding 130 degrees Celsius. Three of the H.Stars points turn out to be exactly collinear, which is evidence that the linear envelope form is the vendor's own model rather than an approximation introduced here.

Stage 4 — Defrost

An air-source heat pump running below roughly 5 degrees Celsius accumulates frost on its outdoor coil and must periodically reverse to clear it, which costs both electricity and heat taken back out of the building's water. This stage scores that penalty across eight source temperatures through and below the frosting band, at two setpoints, two machine sizes and three compressor control types.

Defrost is where air-source heat-pump projects lose energy that a spreadsheet estimate never sees. It costs twice: the compressor runs to melt the frost, and the heat to do it comes out of the building's own hot water. In a cold climate the machine spends thousands of hours in the frosting band, so a model that treats defrost casually will be wrong about the annual bill.

Headline result

All 96 scenarios pass, with 1,272 individual gates scored. Defrost electricity matches EnergyPlus to +0.02 to +0.10 percent and the heat taken back out of the loop to +0.02 to +0.10 percent. Because EnergyPlus meters defrost separately, these are scored directly rather than inferred from a total — the defrost term is verified on its own, not absorbed into a compressor number that happens to come out right.

What we improved in CogenS while running this validation

The defrost accounting was rebuilt during this stage after the first comparison ran up to 19 percent high at the cold end. Three things were wrong and all three are corrected. The heat penalty is additional load the machine must carry, not a reduction in the capacity it has available — the machine's output holds and its electricity climbs, which is the opposite of what a capacity derate produces. The defrost frequency term is a count of cycles per hour, not a fraction of the hour spent defrosting. And the reported defrost load is coupled to the part-load ratio, because the load appears in the denominator of the ratio it depends on; resolving that coupling properly turns it into a quadratic with a closed-form solution. After the rebuild the worst deviation across all 96 scenarios is 0.19 percent, down from 19 percent. A machine that idles at low load no longer pays a full-load defrost bill.

Stage 5 — Auxiliary heater coupling

A heat-pump boiler in a cold climate is nearly always paired with an auxiliary heater that covers what the heat pump cannot. This stage runs the pair in series across seven source temperatures from minus 20 to plus 10 degrees Celsius, three load levels and two setpoints, and checks that the handoff between them happens at the right point and that the two heat streams sum to the load rather than stacking on top of it.

The auxiliary is where a heat-pump boiler project's operating cost is decided. Every kilowatt-hour the heat pump cannot deliver is bought as gas instead, at a different price and a different emissions factor. Getting the handoff point wrong by a few degrees of outdoor air moves the annual gas bill by a large fraction.

Two different reasons the auxiliary runs

A heat pump can fall short in two independent ways, and they are not interchangeable. It can run out of capacity — it cannot make enough kilowatts. Or it can run out of temperature — it has plenty of capacity but the operating envelope caps how hot it can send the water, so it cannot reach setpoint. The auxiliary sits downstream in series and lifts the water the rest of the way, which means the two heat streams add up to the load rather than stacking on top of it. A model that only tracks the capacity shortfall is blind to the second case entirely, and will under-predict gas use on precisely the coldest days.

Headline result

All 42 scenarios pass. The split between heat-pump and auxiliary heat matches EnergyPlus to within 0.07 percent, and auxiliary fuel to 4.96 percent against a 5 percent band. The engine reports which of the two limits bound each hour, so a project can see whether adding capacity or raising the achievable water temperature is what would actually reduce its gas use.