What is interval data?

The mechanics

A traditional meter is a running total. Someone reads it, subtracts last month's number, and the difference is the bill. Nothing in that record says when the electricity was used.

An interval meter writes a value at the close of every interval. A 15-minute meter produces 96 readings a day, 2,880 in a 30-day month and 35,040 in a year. That record is what makes it possible to answer questions a monthly total cannot touch: what set the demand peak, what runs at 2 a.m., whether the weekend shutdown actually shuts down, and whether a project that was supposed to save 6% did.

Three things about the interval matter more than the number of minutes:

• It should match the utility's demand interval. For anything about demand charges, the facility's data has to be aligned to the same interval boundaries the utility bills on. Data at 1-minute resolution that is not aligned to the utility's 15-minute boundaries will produce a different peak from the one on the bill, and the difference is not an improvement.
• Clock alignment across meters is the hard part. Utility validation practice treats a meter within 3 minutes of true time as passing, requires proration between 3 and 75 minutes, and treats anything beyond 75 minutes as data that must be estimated rather than used. Drifting clocks do not break obviously; they quietly corrupt peak attribution and any join between electrical data and production data.
• Higher resolution solves fewer problems than vendors imply. Ten-second data does not make tenant billing legal, does not fix an incorrect circuit-to-asset map, and cannot see a voltage sag. Each of those has a different binding constraint. Resolution addresses only latency and attribution.

Validation, estimation and editing — the part nobody documents

Interval data has gaps. This is normal and every serious utility has a formal procedure for it, usually called VEE: validation, estimation and editing. A monitoring platform that does not implement something equivalent is presenting an incomplete record as a complete one.

The standard checks:

Check	What it catches
Sum check	Interval totals that do not reconcile to the register read
Spike ratio	An interval more than roughly 1.8× its neighbours
High/low bounds	Values outside 50–200% of the expected range for that period
Zero/negative	Unexpected zeros, or negative values without on-site generation

And the standard estimation rules: gaps of two hours or less filled by linear interpolation; longer gaps filled by reference-day averaging from comparable days.

The rule that matters most is procedural rather than technical: every estimated interval must be flagged as estimated, in the interface and in every export. Silent interpolation presented as measurement is the fastest available route to a customer never trusting the system again.

Gaps are not an anomaly to be embarrassed about. A two-year study of a 331-node campus wireless network measured 5–25% packet loss seasonally, with backbone losses over 50% roughly once every six weeks. Gap handling is part of the data product.

Worked example

A 15-minute meter on a packaging line shows 45 kW during production and 45 kW at 3 a.m. on a Sunday. A monthly bill cannot distinguish those two facts; the interval record makes the second one obvious.

Off-hours for a single-shift operation run about 128 hours a week (nights plus weekends).

• 128 h/week × 52 weeks = 6,656 hours a year
• 6,656 h × 45 kW = 299,520 kWh
• At $0.11/kWh: $32,947 a year on one load that nobody turned off

This class of finding — equipment running outside schedule, loads that never turn off, standby power, staging that never sheds — is consistently where the money is in interval data, and it requires no analytics beyond reading the record correctly.

What interval data is not

It is not real-time data. A 15-minute interval is written at the close of the interval and typically delivered hours later, sometimes the next day. For operational response — curtailing on a demand-response call, catching a peak as it forms — a facility needs a separate live feed. Interval data is the billing and analysis record.

It cannot see power quality events. A voltage sag lives at half-cycle resolution, roughly 8 milliseconds, and lasts anywhere from half a cycle to a minute. Averaged over 15 minutes, or over 10 seconds, it disappears entirely. No amount of interval resolution substitutes for a power quality instrument. See what is a voltage sag.

It is not automatically available from the utility. Several large utilities do not release interval data to customers at all, or release it only for accounts above a size threshold, or release it with a multi-day lag. It should be treated as enrichment where it exists, with the bill as the fallback, not as an assumed input.

It is not the same as submetered data. Utility interval data is whole-facility. It says the peak happened; it does not say which three circuits caused it. See what is submetering.

More intervals is not better data. Retaining 10-second data indefinitely multiplies storage cost by 90 against 15-minute data and answers no additional question after the first month. A defensible retention policy is high resolution hot for 7 to 30 days, 1-minute for a year, and 15-minute permanently.