Model Calibration With Bills

Calibration means adjusting a model, within plausible limits, so that its output aligns more closely with measured energy use. It starts from a model built on surveyed inputs, then compares modelled demand with a year of metered bills or readings, and then investigates the gap. Where the gap has a physical explanation, such as a heating system that runs longer than the standard pattern, or a wall whose construction was misrecorded, the corresponding input is corrected. Where no explanation is found, the gap is recorded as unexplained rather than hidden by an invented adjustment.

The key discipline is that calibration adjusts inputs toward evidence, not toward a desired answer. Changing a wall U-value to a value the wall cannot plausibly have, merely to make the model match a bill, produces a model that matches one year and misleads every later decision. A calibrated model should be able to say, for each change made, what evidence justified it and what range of values remains plausible.

In practice, calibration proceeds in stages. First, align the comparison: same fuels, same period, and an allowance for weather differences between the modelled standard year and the actual year. Second, check the easy explanations: occupancy, heating hours, thermostat settings, a change of household, a period of absence, or a fault such as a boiler stuck on. Third, test building explanations: fabric inputs, airtightness, system efficiency and controls. Only after those stages does it make sense to consider finer adjustments, such as internal gains or ventilation rates.

Meter data quality shapes what calibration can achieve. Complete, dated meter readings for each fuel, with meter changes and estimated readings flagged, support a stronger calibration than a handful of bills with gaps. Smart meter data, where available, adds daily and seasonal patterns that can distinguish, for example, high baseload use from heating-driven winter peaks. Without that granularity, calibration can match an annual total while missing the pattern that would identify the cause.

Before calibrating, gather a full year of readings, note occupancy changes, heating settings and any building work during the year, and obtain the model's input schedule. Work through inputs in order of influence, changing only those with an evidence basis, and record each change with its reason. If a change is made purely to improve the match, without evidence, label it as such and treat the calibrated result as provisional.

Also decide what the calibrated model will be used for. A model calibrated to one household's behaviour is useful for understanding that household's use and for testing measures for that household, but it is no longer a standardised comparison with other homes. For a certificate or a compliance purpose, the standardised model remains the official version, and the calibrated version sits alongside it as advice, not as a replacement.

By place, calibration practice is similar in principle across systems, because it is a modelling technique rather than a national rule. In the UK, SAP-based ratings remain standardised regardless of calibration, so a calibrated SAP-style model is an advisory variant, not a new EPC. In EU member states, national methods similarly define the official rating, and calibrated or measured-data analyses complement rather than replace it. The status distinction matters wherever the model is used.

For homes without good meter data, for example where fuels are bought in bulk or where a home changed hands recently, calibration will be weaker, and that weakness should be stated. A calibration based on three months of estimated bills is not comparable with one based on two years of half-hourly data, and the report should say which it is.

A common mistake is to calibrate by tuning a single fudge factor until the annual total matches, and then to present the tuned model as validated. A single-year match can be achieved by many combinations of wrong inputs, and the tuned model will fail on a different year or a different measure. Another mistake is to calibrate to bills that include unrelated energy, such as a workshop, an electric vehicle or a hot tub, without separating those loads first.

A further mistake is to calibrate once and then treat the result as permanent. Occupancy, systems and fabric change, and a calibration ages like any other assessment. Revisit it after major work, a change of household or a system replacement, rather than carrying an old calibration forward silently.

Keep the original model, the meter data, the calibration log with each change and its evidence, and the calibrated result together, with dates. That bundle lets a later professional see what was adjusted, why, and how much uncertainty remains.

Questions to ask include: which inputs were changed on evidence, which remain uncertain, and does this calibrated model still support the decision, or does it only explain last year's bills. Those questions keep calibration honest.