Auditing a Home's Electricity Use: The Night Floor, the Energy Signature and the Sorted List

A bill that doubled with nothing new plugged in

The bill is roughly twice the last one and the house is the same house — same occupants, same appliances, nobody moved in and nothing was bought. The instinct is to go round switching things off at the wall, and it is the wrong instinct, because the money on that document is the product of two independent numbers and only one of them is about the building. Price times quantity. Pulling them apart takes five minutes and no instruments: ignore the money and find the kilowatt-hours, this period's and the same period a year ago, then divide each by the days it covers, because ninety-two days against seventy-eight is eighteen per cent more energy at an identical daily rate. If daily kilowatt-hours are flat and the money doubled, the building is innocent and the fault is commercial — a fixed contract that ended, a standing charge that rose, a discount that expired, a transfer onto a deemed rate after a supplier change. Insulating a house to fix a pricing problem is the most expensive available way to be wrong, and it happens every winter.

If daily consumption genuinely moved, four candidates remain and they leave distinguishable marks. Weather is the commonest by a wide margin: a house with any electric heat can double its draw in a cold spell with not one habit changed. A new large load is second, and it usually announces itself by starting on a date somebody can name. Third is something stuck permanently on — an immersion thermostat welded closed, an underfloor mat energised beneath a rug laid in October, trace heating that never left frost mode. Fourth is the reading itself, where an estimate followed by a true read drops months of catch-up into one document; rule that one out first, because it is the only entry on the list that costs nothing to test.

Weather is not behaviour, and degree days are how you tell them apart

The most useful thing that can be done with a year of consumption data costs nothing, appears on no bill, and takes twenty minutes in a spreadsheet. Take daily kilowatt-hours from the smart meter data or the supplier portal, take daily mean outdoor temperature from the nearest published station, and plot one against the other with temperature on the horizontal axis. What comes back is the house's energy signature, and it is the closest thing a domestic building has to a diagnostic trace.

It has a shape, and the shape is the finding. Above some outdoor temperature the points settle into a horizontal band: the consumption that does not care what the weather does — lighting, cooking, refrigeration, electronics, laundry, hot water, and the standing draw of everything never switched off. Below it, they climb along a line whose steepness is the heating. Intercept and slope are separate quantities with separate remedies, and almost every wasted retrofit pound is spent attacking one while the problem sits in the other. The slope, in kilowatt-hours per degree-day, measures envelope and heat source together, so identical slopes can mean opposite things: a leaky heat-pumped house looks gentler than a tight one on panel heaters, because one returns several kilowatt-hours of heat per kilowatt-hour bought and the other returns exactly one. The nameplate tells you which you have.

The base temperature — where the sloped line meets the flat band — is a property of the specific building rather than a convention, and this is where published degree-day series quietly mislead. North American series default to a base of 65 degrees Fahrenheit and British ones to 15.5 degrees Celsius, both inherited from buildings losing far more heat than a modern one does. CIBSE TM41 sets out how base temperature is chosen and what the analysis does when it is chosen badly; the ASHRAE Handbook chapter on energy estimating and modeling methods gives the degree-day and bin methods the same treatment from the design side. In practice the procedure is empirical: fit at several candidate bases and keep the one that straightens the line, because that temperature is telling you something real about internal gains and insulation.

Two cautions before the result is used. A house with air conditioning has two slopes and comes back as a V with a flat floor between the arms, most clearly where a heat pump runs in both directions. And a regression is only a baseline if it fits: ASHRAE Guideline 14 and the IPMVP both set out how a baseline model is judged before savings are calculated against it, and twelve monthly totals make a far weaker instrument than three hundred and sixty-five daily points.

The night floor, and how much of the year hides inside it

Now go the other way and look at the smallest number in the dataset. In a half-hourly series, find the lowest consumption recorded between roughly two and five in the morning across a normal week. That is what the house draws with everybody asleep and nothing being used on purpose, and it runs for all 8,760 hours of the year whether anybody is in the building or not. Double the half-hourly figure to get average kilowatts, multiply by 8,760, and in a house that does not heat, drive or make hot water on electricity the annual total that falls out is regularly the largest single line on the whole audit — and the one nobody has ever seen, because it never draws attention to itself.

The arithmetic is brutally simple, which is exactly why it gets ignored: one watt running continuously is 8.76 kilowatt-hours a year. Forty-five watts of always-on draw — a router, an optical network terminal, two set-top boxes, a console in rest mode, a doorbell transformer, a couple of power supplies warm to the touch — is 394 kilowatt-hours a year, which beats the kettle comfortably. But the floor is not standby alone, and treating it as though it were is how the diagnosis goes wrong. Refrigeration compressors cycle through the night; a circulator left on a permanent setting holds a steady few dozen watts; a cellar dehumidifier, a sump pump, an inverter's tare draw, a car charger idling with nothing plugged in and an outbuilding on the same supply all live in that floor, and every one is a different repair.

Where no interval data exists, the floor can be measured at the meter, and the technique is worth knowing because it also yields per-circuit numbers no supplier portal can. Nearly every electronic meter has a pulsing indicator on the faceplate with a constant printed beside it in impulses per kilowatt-hour, commonly 1,000 imp/kWh. Time the interval between pulses and the instantaneous load falls straight out: watts equals 3,600,000 divided by the product of the meter constant and the seconds between pulses. At 1,000 imp/kWh, one pulse every thirty-six seconds is exactly 100 watts and one every 3.6 seconds is a kilowatt. It is a true active-power reading from a certified instrument, which is more than can be said for most of what gets clipped around a cable.

1. Pick a night nobody needs anything: no laundry, no delay-timed dishwasher, no car charging, no immersion boost scheduled.
2. Time ten pulses rather than one and divide, so a single mistimed press does not become the answer.
3. Take the whole-house floor first, then switch off one circuit at a time and re-time, writing each difference against the circuit label.
4. Do the socket circuits last and slowest — small always-on loads only separate out by unplugging them one at a time with the circuit back on.
5. Leave alarm, freezer and medical circuits energised, then restore everything and write the per-circuit watts inside the consumer unit door.

Nameplate lies, and a clamp meter lies differently

A nameplate rating is a ceiling declared under the manufacturer's stated conditions, not a description of ordinary life. Anything with a compressor, a thermostat, a variable-speed motor or an inverter drive spends most of its running time well below the number moulded into its label; anything with a heating element spends most of its time at exactly that number or at zero, with nothing in between. A census built from nameplates puts the kettle, the dryer and the oven at the top and leaves the actual answer off the page.

For anything on a plug, a socket-mounted energy monitor settles it, and the specification worth reading is its accuracy class rather than its display resolution. Domestic metering accuracy is defined by IEC 62053-21 for static active-energy meters in classes 1 and 2, by ANSI C12.20 for the tighter revenue classes, and by Annex V of the Measuring Instruments Directive 2014/32/EU for the class A, B and C designations on European equipment. The consequence sits at the bottom of the range: a monitor honest about a two-kilowatt load can be badly wrong about a three-watt one, which is unfortunate given that three-watt loads running all year are precisely what the audit hunts. Gang several small devices onto one monitored extension lead so the total lands where the instrument can resolve it.

Clamp ammeters are the instrument most likely to produce a confidently wrong number. A clamp measures RMS current in one conductor. Multiplying by nominal voltage gives apparent power in volt-amperes, and the active power the utility meter records and bills is that figure times the power factor — well below unity for a small single-phase induction motor under part load, and degraded further for switch-mode supplies by a current waveform that is not remotely sinusoidal, which a non-true-RMS clamp then gets wrong as well. This also disposes of a product that sells steadily online: a domestic tariff bills active energy in kilowatt-hours, reactive power is not on the bill at all, and no plug-in power-factor device can reduce a household electricity account.

Then duty cycle, which is the difference between a measurement and a number. A refrigerator observed for five minutes reads either ninety watts or zero depending on when you looked. Leave the monitor on it for a full twenty-four hours and read the accumulated kilowatt-hours; that is the only figure worth writing down.

Reconciling the census against the meter, because the gap is the finding

Add the sorted list up and compare the total against the annual kilowatt-hours the meter recorded. The two will not match, and the difference is not an embarrassment to be smoothed over — it is the most informative number the audit produces, because it measures directly how much of the house has not yet been found. A census accounting for sixty per cent of the meter has not identified the biggest load; it has identified the biggest load among the ones that were easy to measure, which is a different and much weaker statement.

This is not a nicety invented here. Reconciling a disaggregated model against measured consumption is what separates an energy audit from a walk round with a clipboard, and it is written into the standards that govern the work: BS EN 16247-1 with its buildings-specific Part 2, ISO 50002, and the residential ANSI/BPI-1100-T home energy auditing standard all require the analysis to be tied back to metered data rather than left as an unchecked sum of assumptions. CIBSE TM22 sets out the same discipline, breaking consumption into an accounted tree and treating whatever will not fit as an explicit residual, and ASHRAE Standard 211 formalises it into audit levels on the commercial side.

When the residual is large the causes repeat. Hard-wired loads get missed because a plug-in monitor cannot reach them — towel rails, humidistat extract fans, underfloor mats, trace heating on a pipe run, an outside light on a failed photocell, a garage or annexe on the same supply. Heating gets underestimated because equivalent full-load hours were guessed. Occasional heavy use gets left out entirely: a workshop compressor, a kiln, a welder, a dehumidifier run for a fortnight after a leak. And sometimes the residual is a fault, in which case it is flat, present at three in the morning, and already visible in the night floor — which is why that measurement comes before the census rather than after it.

1. Subtract the sorted list's total from the meter's twelve-month total, and record the residual as a figure and as a percentage.
2. Split it by season using the energy signature: a residual that grows with cold weather is unmeasured heating, a flat one is a permanent load.
3. Walk the consumer unit and list every circuit with no census line against it — that is where hard-wired loads hide.
4. Re-measure whatever was estimated rather than monitored, starting with the estimate large enough to close the gap on its own.
5. Stop when the residual is smaller than the item you are about to buy, and write down what it was rather than quietly dropping it.

Aiming the next spend, and proving afterwards that it landed

With a sorted list and a reconciled residual the sequence picks itself, in order of increasing cost. Eliminate first: the failed thermostat, the mat under the rug, the timer nobody reset after the clocks changed, the second freezer in the garage holding four things. Reduce second — schedules, controls, setbacks, and the always-on devices that turn out to have a genuine off state. Substitute third, where resistance heating gives way to a heat pump and the coefficient of performance does the work. Generate last, because self-consumed generation is worth only what the consumption it displaces was worth, and generating into a load you had not yet reduced is paying twice for the same kilowatt-hour.

Comparing measures needs one caution. Simple payback is the natural ranking and it quietly favours short-lived measures: a control change paying back in a year and lasting five is not obviously better than a fabric measure paying back in nine and lasting forty, and the payback figures alone say it is. Rank by payback within a category and by lifetime saving across categories. Where a grant, rebate or tax credit exists, read its terms before buying — many schemes require the application, the assessment or an approved installer to be in place before work starts and cannot be applied to an invoice after the fact.

Then prove it on the instrument that found the problem. Re-plot the energy signature and compare slopes and intercepts rather than bills: a fabric or heating measure should rotate the sloped arm, a base-load measure should drop the flat band and leave the slope alone, and a measure that moved neither did not work whatever the next bill says. That weather normalisation is the whole point of the IPMVP's routine adjustments under its whole-facility Option C and of the baseline criteria in ASHRAE Guideline 14 — savings claimed by subtracting one winter's bill from another's are a statement about the weather, and a mild January will happily award full marks to work that achieved nothing.

1. Fix everything on the eliminate list first, then re-read the night floor to confirm each one actually came off.
2. Rank what remains by annual kilowatt-hours saved, price it, then re-rank by lifetime saving rather than by payback alone.
3. Check grant and rebate eligibility, and the order of operations each scheme demands, before placing any order.
4. Keep to one measure per period where the data allows, so the signature can attribute the change to something specific.
5. Leave the record behind: the night floor with its date, the per-circuit watts, the base temperature the regression settled on, and the grid factor and year used for the carbon column.