Interpolating Missing Energy Data: 5 Methods Compared
How missing meter readings become auditable monthly figures without guesswork — the five interpolation methods of the EDM Toolbox at a glance.
Between two readings: the unsolved problem of energy accounting
Meter readings are rarely captured exactly on the last day of the month. The reader visits on the 3rd, the next value arrives on the 27th of the following month. Yet energy accounting, the ISO-50001 report and CSRD evidence all require clean calendar-month figures — otherwise KPIs, baselines and year-over-year comparisons drift.
In practice, allocation is often done manually in Excel: consumption divided by days, multiplied by calendar-month days, done. That works for a linear base load. For heating, cooling or usage-driven load profiles it is simply wrong — and unauditable.
> The total consumption between two readings stays exactly the same — nothing is added or lost, only redistributed by a defined pattern.
Two layers: monthly value and daily view
Before the methods come into play, one distinction matters. The EDM Toolbox cleanly separates two layers:
- **Monthly value:** The authoritative, accounting-relevant figure. It is built directly from the period allocation of the actual readings and does not require any daily interpolation.
- **Daily view:** An additional visualisation layer in the drilldown chart. Only when a user opens it, the period between two readings is distributed across individual days.
Daily interpolation adds no new measurement information — it distributes the measured total according to a defined pattern. For accounting, the monthly value remains authoritative, regardless of how many readings fall within the month.
The five methods at a glance
Each method defines the daily weight differently. Confidence reflects the reliability of the interpolated value.
| Method | Daily weight | Data source | Use case | Confidence |
|---|---|---|---|---|
| **Linear** | constant = 1 | Calendar days | Base load, robust fallback | 90 % |
| **Historical** | Weekday × month | Prior-year profile (≥ 6 months) + holidays | Recurring, usage-driven load profiles | 85 % |
| **HDD-based** | Heating-degree day | Site weather data (Open-Meteo) | Heating, gas, district heat, oil | 92 % |
| **CDD-based** | Cooling-degree day × solar factor | Weather data including global radiation | Cooling, air conditioning | 92 % |
| **Auto** | see below | depends on data | Mixed portfolios, recommended default | Best-fit |
The overall formula is identical for all methods: consumption per day = ( daily weight / sum of all daily weights ) × period consumption. Only the definition of the weight differs.
Weather and standards: HDD and CDD
Both weather-based methods rely on degree-day figures — and those are normatively defined. The EDM Toolbox supports three calculation methods so that Austrian and German practice can both be met precisely:
- **Austria — ÖNORM B 8135:** HGT = (20 − daily mean) for days below 12 °C. Heating limit 12 °C, indoor reference 20 °C.
- **Germany — VDI 3807:** GTZ = (20 − daily mean) for days below 15 °C. Heating limit 15 °C, indoor reference 20 °C.
- **International — Base-N:** HDD = max(0, T_base − T_a) with configurable base temperature (default 15 °C).
Cooling-degree days are defined symmetrically: CDD = max(0, T_a − T_base) with a default base of 18 °C. On top, solar radiation modulates the cooling weight — relevant for glass-heavy buildings where the sun drives cooling demand independently of air temperature:
`CDD' = CDD × ( 1 + 0.25 × min( 1, Solar / 8000 ) )`
The uplift is deliberately capped at 25 % and can never reverse the ranking of two days. Temperature remains the primary driver. If weather data is missing, the method automatically falls back to Linear — and the affected value is honestly flagged as linear rather than faking a weather weighting.
Balance point: the building-specific base temperature
Instead of a fixed heating or cooling threshold, the EDM Toolbox can determine the base temperature empirically from actual consumption. The cleaned monthly consumption is regressed against degree-day sums at various candidate base temperatures; the temperature with the best coefficient of determination (R²) marks the physical switching point between weather-independent base load and temperature-driven share.
The method uses only clean monthly reading pairs, requires at least 12 usable months with over 90 percent weather coverage, and checks for seasonal spread. The result is not a blindly assigned value but the full R²-over-base-temperature curve — transparent and reviewable.
Holiday calendar for the historical method
On public holidays, building occupancy follows the weekend rather than the weekday pattern. The historical method takes this into account: a public holiday is weighted like a Sunday — regardless of the actual weekday.
The applicable calendar is resolved per site (site country → organisation default → fallback AT). Movable holidays are calculated exactly from Easter each year. Austria is fully covered with 13 national holidays. Germany deliberately uses only the 9 federally uniform days — without state-specific ones such as Corpus Christi or All Saints, because without state assignment at the site a fixed subset would falsely apply them to sites in states where the day does not exist.
Auto mode: the rule-based cascade
For mixed portfolios, manually choosing a method per metering point is impractical. Auto mode inspects the data situation per metering point and picks the first applicable, most reliable method:
1. **Weather-based:** If weather data is available and the metering point is temperature-driven (heat, gas, cooling), HDD or CDD applies.
2. **Historical:** No usable weather signal, but at least 6 months of prior-year data — the historical method weights by the individual prior-year profile.
3. **Linear:** Neither weather nor sufficient historical data? Even distribution as robust fallback. Always available.
Every value stays flagged with the method actually used — the auto choice is not a black box.
Confidence model and audit trail
Audits ask not only for numbers, but for their origin. The EDM Toolbox assigns a quality level to every monthly value and thereby makes the estimated share of a report transparent:
- **Measured** — reading at end of month (± 2 days). Confidence 100 %.
- **Estimated** — consumption allocated from a reading period. Confidence 80 to 95 %.
- **Interpolated** — data gap bridged by a helper method. Confidence 40 to 70 %.
- **Forecast** — extrapolation for the current month. Confidence 50 to 80 %.
Every reading, every method and every confidence figure remain traceable and exportable via the audit trail. If an entire month is missing, a gap treatment applies — prior-year comparison, neighbour-month average or weather correlation — visually highlighted and with reduced confidence.
Conclusion
Energy accounting without interpolation does not exist in practice. The real question is not whether but how transparently and how normatively it is done. Five defined methods, a rule-based auto mode, clean references to ÖNORM and VDI, and a continuous audit trail are what make the difference between a robust ISO-50001 report and an Excel guess that surfaces in the next audit.
The EDM Toolbox delivers this model as part of its energy accounting layer — including weather and holiday data, empirical balance-point detection and a visible estimated share in every report.