What to measure, and how often

The question always arrives in the same shape: how many meters do we need. It is the right question at the wrong moment, because the number is not settled by counting panels. It is settled by deciding first which sentences you want to be able to say at the end, and then finding the fewest measurements that support them.

Measuring everything is not a plan

Instrumenting every outgoing way of every panel is expensive, slow and — less obviously — worse for the analysis: a hundred series nobody looks at bury the ten that matter, and when the first one stops working, nobody notices.

The opposite criterion — one main meter and nothing else — has the mirror-image flaw. It says how much was consumed, not where. Enough to pay the invoice, not enough for any decision.

The plan sits in between, and it is shorter than people expect. On almost every site a handful of loads explains the large majority of consumption, and those are the ones to measure first.

The hierarchy, in three levels

Level 1 — the supply point. One per vector: electricity, gas, heat, water, compressed air if it is produced on site. This is the level that has to reconcile with the invoice, and it is the consistency check for everything else.

Level 2 — the large loads. The ones that on their own account for more than a few per cent: the chiller plant, air compression, furnaces, pumping, the main line. Five or ten points, not fifty.

Level 3 — whatever answers one specific question. A machine under observation, a department in dispute, a measure to be verified. This level is born with the question and dies with it, and it is normal for it to change.

The interval is chosen per vector

There is no single correct interval. There is a correct interval for each use of the data.

Two different intervals on the same site are not a problem, provided they are declared. The problem starts when an hourly value and a per-second value end up on the same chart with no way to tell which is which.

Every reading carries four things

A usable measurement is not a number. It is a number plus the context that lets someone reuse it two years from now, when whoever configured it has moved on:

• When, with the time zone stated. Daylight saving moves the day boundary twice a year, and a daily consumption computed over a 23-hour day is wrong silently.
• What, with the unit. kWh and kW are not the same quantity, and confusing the two is the single most common error there is.
• From where, with the identity of the measurement point, not just the device name. Devices get replaced; the measurement point stays.
• How far it can be trusted: read directly, reconstructed, or a declared gap. An honest gap is worth more than an invented zero.

The last one is usually the one missing, and it is the one an audit looks at first.

When estimating is the right call

Estimating is not forbidden: estimating without saying so is. There are points where direct measurement costs more than it is worth — a load carrying one per cent, a branch that would have to be interrupted to fit a meter — and for those a documented estimate is a reasonable choice.

Three conditions apply: the method is written down, the result is marked as estimated everywhere it appears, and the estimate does not cover the items at the top of the ranking. If the largest load on the site is estimated, no analysis built on it will hold.

A minimum configuration that works

For an average site, the starting point that is rarely wrong:

• One meter per vector at the supply point, quarter-hourly.
• Five to ten sub-loads chosen by weight, at the same interval.
• The boundary conditions that explain the variability: outside temperature, and the production variable if there is one.
• A written measurement plan: what, where, at what interval, in what unit, and since when.

That last line is the one that ages best. In two years nobody will remember why that meter was put there, and the document explaining it is worth as much as the meter.