Skip to content
allanninal.dev/build
Collaborate

Part 2 of 7 · Utility meter reader series ~5 min read

How a reading becomes a number you can use

Meter data looks like the most trustworthy data a business has, and it arrives with several ways of being quietly wrong that produce confident-looking charts of nothing.

Key takeaways

  • Every value is actual, estimated or derived, and the three are never mixed in a comparison.
  • A catch-up after estimates should be spread back, and shown as spread.
  • Half-hourly data finds baseload; monthly data cannot.
  • Meter replacements reset the register and need explicit handling.
  • Always divide by days in the period before comparing anything.

The catch-up

Four estimated months followed by a catch-up on the actual readingA stacked bar chart with five bars in kilowatt-hours. Two series: consumption billed in green, and catch-up from four under-estimates in red. November, December, January and February are each estimated at between four thousand two hundred and four thousand three hundred with no catch-up. March, an actual reading, shows four thousand four hundred billed plus a catch-up of five thousand six hundred. A note says nothing happened in March, and the red band belongs to November through February.05000100001500020000~4200Nov (est)~4200Dec (est)~4300Jan (est)~4300Feb (est)~10000Mar (actual)Consumption billed, kWhCatch-up from four under-estimatesNothing happened in March. The red band belongs to November through February.
Fig 1. Five months of billing where four were estimated. The March bar is the one somebody investigates and it is the only one that is correct.

The system’s job here is not to fix the billing; it is to stop the chart lying. When an actual reading follows estimates, the difference is spread back across the estimated periods proportionally, and every one of those periods is marked as derived rather than measured.

That produces a series that is approximately right everywhere instead of exactly wrong in one place, and it stops the recurring investigation into a month where nothing happened.

Three kinds of value

Actual is a real meter reading. Estimated is the supplier’s guess. Derived is what the system computed by spreading a catch-up back. All three are useful and mixing them silently is what produces nonsense.

The rule that follows: any comparison between periods states how much of each period was actual. “January against last January” is a different claim when one is fully metered and the other is eighty per cent estimate, and saying so takes one line.

Resolution decides what you can find

Data availableWhat you can seeWhat you cannot
Annual billWhether it went upAnything actionable
Monthly billSeasonal shape, big changesBaseload, schedules, spikes
Daily readingWeekends against weekdays, shutdownsWhat runs overnight
Half-hourlyBaseload, start-up, timers, peak demandWhich circuit it is
Sub-meteredWhich circuit or areaWhich individual machine

The jump from monthly to half-hourly is where almost all the value is, and for electricity it is usually available already: most commercial supplies are half-hourly metered and the data can be requested from the supplier or the data collector at no cost.

That request is frequently the single highest-value action in an energy project, and it is an email. A business paying for half-hourly metering and looking at monthly bills is the common case.

Meters that change

How a meter reading is validated and normalisedA vertical chain of five steps entered by a box labelled A new reading from a meter. Step one asks whether it is lower than the last, since registers go up; if so it exits to Meter change or rollover, and never records a negative period. Step two asks whether the meter serial is the same, which appears on the bill; if not it exits to New meter, closing the old series. Step three computes the period length in days between readings. Step four computes a per-day figure, the only comparable one. Step five stores it with its type: actual, estimated or derived. A note says a meter change mid-year is the most common cause of an impossible-looking series.AWS ACCOUNTA new readingfrom a meterLower than the last?registers go upMeter change or rollovernever a negative periodyesSame meter serial?it is on the billNew meterclose the old seriesnoPeriod lengthdays between readingsPer-day figurethe only comparable oneStored with its typeactual, estimated, derivedA meter change mid-year is the most common cause of an impossible-looking series.
Fig 2. How a raw reading becomes a comparable number. The first two gates catch the discontinuities that otherwise produce enormous or negative consumption.
  • Compute
  • Database
  • App integration
  • Management
  • Analytics

Per day, always

The cheapest correction available and the most frequently omitted. A 28-day February against a 31-day January is ten per cent lower before anything else is considered, and a quarterly bill covering 89 days against one covering 93 is four per cent.

Everything downstream works in units per day, and the totals are reconstructed only for reporting the actual bill. It removes an entire class of false movement at no cost.

Next: what is running when nothing should be.

All posts