What we found
- A hybrid plant is the cheapest way to serve this load. About 25 kW of heat pump, 500 L of storage and a 20 kW element costs over 15 years. The best plant without an element (40 kW, 1,500 L) costs .
- It wins on equipment, not efficiency. It uses about 2.5% more electricity, but its equipment costs 43% less. The element covers the peak at a fraction of the cost of heat pump capacity.
- It wins by more when real demand varies. On days heavier than the scheme profile, plants without an element fail to hold 45 °C unless they grow to 50 to 80 kW and 2,000 to 3,000 L. The hybrid only needs a bigger element, which is cheap.
- Three things can reverse it: an all-in element price near A$2,000/kW, a demand charge near A$45/kW/month, or an element thermostat deadband near 12 K. Tariff, discount rate, heat pump price, climate and building size change how much the hybrid saves, never whether it saves.
- The sizing rule is simple. Per 1,000 MJ/day: 25 kW of heat pump, 500 L, and 20 kW of element, rising to 45 to 60 kW where demand varies a lot. Set the element setpoint at 45 °C plus its deadband.
- TRNSYS agrees on the cost gap and on delivery. It also found a thermostat layout problem in our first setup, which we have corrected.
The question
Given a building with a fixed daily load and load profile, which combination of heat pump capacity, element capacity and storage delivers the load at the lowest 15-year spend?
The usual instinct is to size the heat pump to carry the whole load, because it is the efficient part of the plant. Our hypothesis was that a smaller heat pump plus an electric element would serve the same load for less money, because heat pump capacity is expensive and the element only needs to cover the peak.
The answer
Hybrid
Best without an element
The cheapest hybrid on the full grid uses an even smaller 250 L store (), but at 23 compressor starts a day it sits right on our cycling limit. The 500 L design costs 1% more and starts the compressor half as often, so we use it as the reference. Everything from 250 to 600 L is within 2% of the minimum.
With interest on money
The headline is undiscounted. Discounting shrinks future bills relative to day-one equipment, which favours the plant that spends less up front.
Why the hybrid wins
Equipment cost rises steeply with heat pump capacity, while electricity hardly moves anywhere between 25 and 120 kW. Capacity is what costs money.
A smaller, cooler store loses less heat and lifts the heat pump's COP. The element's COP of 1 costs slightly more than both gains together.
The element holds the top of the store at its cut-in temperature. Set it at 50 °C and delivery never falls below 45 °C.
Capacity is what costs money
Same electricity, different mix
The element holds the delivery floor
In every winning hybrid the lowest delivery temperature equals the element's cut-in: its setpoint minus the 5 K deadband. That makes "passes 45 °C" a choice of setpoint, provided the element is large enough to hold it on the worst morning. It also explains why the cheapest design has no margin: it puts the cut-in exactly on the requirement. A 5 K margin (setpoint 55 °C) costs about A$2,200 over 15 years.
Storage buys fewer compressor starts, not energy
The whole design space
Every cell is the cheapest design at that heat pump size and storage volume that passes every rule. The two outlined cells are the cheapest hybrid and the cheapest plant without an element. Hover for values.
What can change the answer
We re-priced the same simulations under every cost assumption, and re-simulated the plant under every control, hardware, climate and load change. Each test re-optimises both the hybrid and the plant without an element.
| Factor | Range tested | Hybrid saving | Can it reverse the result? |
|---|---|---|---|
| Element all-in price | A$44 to 4,000 per kW | 16% to −11% | Yes near A$2,000/kW |
| Demand charge | A$0 to 60 per kW per month | 16% to −5% | Yes near A$45, or A$65 at 7% discount |
| Element deadband | 2 to 12 K | 15% to 0% | Yes near 12 K |
| Draw shape | flat day to burst day | 5% to 27% | Nearly on a flat day |
| Demand variance | none to 30% day to day | 16% to 38% | No it raises the saving |
| Electricity price | A$0.10 to 0.60/kWh | 28% to 6% | No |
| Discount rate | 0 to 10% | 16% to 28% | No |
| Heat pump price, equipment discount | A$779 to 1,600/kW; 70 to 115% of list | 12% to 22% | No |
| Climate, load size, COP, pipe length, heat pump deadband | 7 zones; 500 to 4,000 MJ/day | 10% to 16% | No |
| Element position, compressor timers, tank loss, vessel size, certificates, sanitation | within 2 points | No |
Cost assumptions, one at a time
The two prices that decide it
Element price and demand charge are the only costs that can each reverse the result on their own, and they compound. The map shows the hybrid saving for every combination: orange where the hybrid is cheaper, blue where a plant without an element is, and a heavy line at break-even.
Real days are not the scheme day
The scheme test uses one repeating day. Real buildings have heavy days. We ran 312 designs on four seeded years of varying demand each, and kept only the designs that hold 45 °C on every day of every year.
Where it holds
Two results stand out. A flat daily draw leaves little peak for the element to shave, and the saving drops to 5%. A wide element deadband lowers the delivery floor one-for-one, and at 12 K the hybrid needs so high a setpoint that it loses its advantage. The burst day, half the daily volume in three morning hours, passes nothing on the mini-grid; on a wider grid the hybrid is 27% cheaper, but with a 90 kW element on a 60 kW heat pump.
Does TRNSYS agree?
Four designs were rendered into the TRNSYS 15 and TRNSYS 18 template libraries and run through the production harness at the standard 0.02 h timestep.
| Design | Engine | Lowest delivery | Savings | Electricity | Element |
|---|---|---|---|---|---|
| Hybrid 25 kW, 500 L, 20 kW element | Python | 45.02 °C | 66.8% | 115.3 GJ | 13.3 GJ |
| TRNSYS 15 | 45.07 °C | 66.3% | 116.9 GJ | 12.3 GJ | |
| TRNSYS 18 | 45.1 °C | 66.3% | 116.8 GJ | 12.4 GJ | |
| Without an element 40 kW, 1,500 L | Python | 48.99 °C | 67.6% | 112.5 GJ | 0 |
| TRNSYS 15 | 47.71 °C | 66.7% | 115.3 GJ | 0 | |
| TRNSYS 18 | 48.3 °C | 66.8% | 115.3 GJ | 0 |
- The cost gap is the same. 15.9% in TRNSYS 15, 16.0% in TRNSYS 18, 15.6% in the Python engine.
- The engine does not over-credit the element. It reports 3 to 9% more element energy than TRNSYS. TRNSYS uses 1.3 to 2.5% more electricity overall, which traces to its own heat pump loop.
- The thermostat position matters. With the element thermostat above the element, TRNSYS cannot run the hybrid at 20 tank nodes. With it below, everything converges and the hybrid holds 45 °C up to about 1,300 MJ/day.
- The plant without an element is marginal at 20 nodes. TRNSYS 15 fails it at 44.5 °C, which would make the hybrid's advantage larger in the production template, not smaller.
What we corrected on the way
| What we first thought | What the tests showed |
|---|---|
| The 60% savings rule limits how small the heat pump can go. | Not at the optimum. Heat pump size is an economic balance; the rule only binds when electricity is cheap. |
| Thermostat one node above the element, the textbook layout. | This layout stalls the heater routine in TRNSYS and the engine. We moved it below the element; the cost answer did not change. |
| Designs below 500 L could not be simulated. | That was the thermostat layout. They run, and the 250 L design is real, but it sits on the compressor cycling limit. |
| Sanitation would limit small heat pumps. | It never binds, at any size. |
| Two hard rules, compressor cycling and sanitation, were enough. | Element cycling matters too. Elements of 45 kW and above run 3 to 4 minute bursts about six times a day. |
What is not settled, and next steps
- The all-in installed price of a staged element with its control gear.
- The demand-charge structure the audience actually pays.
- Element staging. The engine has one element at one capacity; large elements would really be several stages, which would also fix their short bursts.
- Size-dependent heat pump efficiency. One performance map is scaled to every size.
- A TRNSYS check of the designs that survive variable demand, and of New Zealand climates.
- Installation cost, excluded by decision.
- Get an electrical contractor's price for a 30 kW and a 60 kW staged element, installed.
- Pick a representative NZ commercial tariff and its demand charge.
- Add a staged element to the engine and re-run the variable-demand tests.
- Run a variance-proof hybrid in TRNSYS at 10 and 20 nodes.
- Present the method and the break-even map, rather than any single design, as the result.
Basis and method
Engine. The EnergyAE Python commercial hot water engine, pinned at one version, with a full-year simulation for every design at a 0.02 h timestep. About 4% of designs needed a finer timestep (0.01 or 0.005 h) to converge; each result records which it used. Spot checks show the finer timestep changes electricity by 0.3% on average.
Plant model. One R290 heat pump with a normalised performance map scaled to each capacity, a fixed 5 K temperature rise across the heat pump, parallel vessels of up to 1,000 L, an immersion element at 25% of vessel volume from the top with its thermostat one node below, the element allowed to run alongside the heat pump, a heat pump setpoint of 65 °C with a 5 K deadband, a weekly sanitation cycle, and compressor timers of 10 minutes on and 5 off.
Costs. Equipment at Australian supplier list prices compiled in 2026, supply only, ex GST: heat pump A$965/kW (range A$779 to 1,150), storage vessels A$7.97/L up to 1,000 L each, auxiliary equipment 22% of heat pump and vessels, immersion element A$44/kW. Installation excluded. Electricity A$0.20/kWh flat. Certificates are not in the headline; at a fixed registered load they are worth about A$14k for every design and do not change the ranking.
Variable demand. The engine's own generator perturbs each day's volume (lognormal, correlated day to day) and timing, keeping the annual energy unchanged. Four seeded years per design.
Record. Every test, in the order it was run, with its result and what it taught us, is kept in the EnergyAE simulation log for this study.