The model versus the field: 3.53 against 3.52
My heat pump model predicts a seasonal efficiency of 3.53; Dutch field monitoring across six thousand homes measures 3.52. But the validation also produced an uncomfortable discovery: not a single Dutch showcase project publishes a measured self-sufficiency figure.
The test I had been putting off
Image above: AI impression, not a construction drawing.
A model can produce beautiful numbers for months without anyone holding them against reality. That was true of mine. So I set the lab's canonical results against the best available field measurements, under one agreement: a deviation is worth more than a reassurance.
The most exciting number first. My heat pump model arrives at a seasonal efficiency of 3.53. The most recent Dutch field monitoring, over six thousand homes with smart-meter data, measures an average of 3.52 for all-electric space heating. That is almost exactly on target, and the honest story behind it makes it better: my model first sat at 3.78, until an adversarial review (by a competing AI model, of all things) pointed out a missing defrost correction. Without that correction the model would have fitted the small floor-heating-only subgroup, not the population average.
Where the model is too pretty
The same field monitoring measures domestic hot water separately, and there the figure is 2.3. My model computes hot water through the heat pump curve at 55 degrees and knows no boiler or standby losses. So there I am probably too optimistic, by an estimated one to two percentage points of self-sufficiency. That is revision candidate number one, and it is now marked as such in the assumptions register.
Nobody measures what everybody claims
The strangest result of the validation is negative: not a single Dutch showcase project publishes a measured self-sufficiency figure. The Aardehuizen earthship village generates 130% of its consumption. Ecovillages call themselves energy-positive. German villages reach 500% generation. All of these are annual balances: summed over the year, they generate more than they use.
Self-sufficient is something else. The only project where autarky was actually measured, a neighbourhood of 39 all-electric rental homes in Uden without a battery, comes out at 30%. My model neighbourhood without seasonal storage sits between 45 and 58%, with a battery. Those numbers get along fine. But "140% generation" and "45% self-sufficient" describe the same kind of village, and almost nobody explains the difference. Summer produces the surplus; the winter import remains.
For the water layer, the cleanest test came from Flanders, where rainwater tanks are mandatory and mass data therefore exists: households with a tank cover 25 to 30% of their use. My model says 24 to 26%. Inside the band, without tuning.
What this rests on
- This is validation against bands and averages, not time series. The real test, hourly data from a real project next to a prediction from this model, has not been done yet.
- Schoonschip in Amsterdam has that data, at quarter-hour resolution, but does not publish it; it sits with the measurement parties and the cooperative.
- At the Aardehuizen, the first full measurement year of their new community battery is running right now. That is the test this model can lose, which is exactly why I want to do it: deposit the prediction first, put the measured data next to it afterwards. If anyone from that project reads this: the offer stands.
- The field monitoring itself has small subgroups for strict new-build; the population average is the best available reference, not a perfect one.