IPMVP options A to D, explained for building automation data

Energy savings cannot be measured directly. A saving is energy that was not used — so it can only be calculated by comparing what happened with an estimate of what would have happened without the change. How that estimate is built decides whether a savings claim survives scrutiny.

The most widely used framework for doing it is the International Performance Measurement and Verification Protocol (IPMVP), published by the Efficiency Valuation Organization (EVO). It defines four options, A to D. This guide explains each, when it fits, and what it needs from your building automation data.

The core idea: baseline, reporting period, adjustments

Every IPMVP option rests on the same equation:

Savings = (Baseline-period energy − Reporting-period energy) ± Adjustments

• Baseline period — the time before the efficiency measure, used to learn how the building or system normally uses energy.
• Reporting period — the time after the measure, when savings are claimed.
• Routine adjustments — corrections for things expected to vary, such as weather, occupancy or production. Usually built into a model using these “independent variables”.
• Non-routine adjustments — corrections for one-off changes unrelated to the measure: a new floor fitted out, equipment added, operating hours extended.

Without adjustments, a mild winter can look like a successful retrofit — and a cold one can hide real savings.

Option A: retrofit isolation, key parameter measurement

Option A draws a boundary around the equipment affected by the measure and measures only the key parameter, estimating the rest. In a lighting retrofit, for example, the fixtures’ power draw might be measured while operating hours are estimated from schedules.

It is inexpensive and suits simple, predictable measures. The weakness is the estimate: it must be justified and documented, and the parties relying on the result must accept it.

Option B: retrofit isolation, all parameter measurement

Option B uses the same boundary but measures everything needed to calculate energy use, usually continuously. A new variable-speed drive on a pump, or a chiller replacement with its own metering, are typical examples.

This is where a building automation system shines. Trended power, flow, temperatures and runtime from a Niagara station are exactly the data Option B needs — provided they are collected at the right interval and kept.

Option C: whole facility

Option C measures savings at the utility meter or a whole-building meter, usually with a regression model relating energy use to weather and other variables. It captures the combined effect of several measures and their interactions.

The catch is noise. A whole building’s consumption varies for many reasons, so small savings can disappear into normal variation. A common rule of thumb is that expected savings should be a meaningful share of total use — roughly ten percent or more — for Option C to detect them reliably.

Option D: calibrated simulation

Option D uses a computer simulation of the building, calibrated against measured data, to estimate what energy use would have been. It is used when baseline data does not exist — new construction, for instance — or when many interacting measures make other options impractical.

It is powerful, but only as good as its calibration, and it requires specialist modelling skills. ASHRAE Guideline 14 publishes statistical criteria commonly used to judge whether a model is calibrated well enough.

The four options side by side

Choosing an option

The right option depends on a handful of questions:

• How large are the savings compared with total use? Small savings favour A or B; large, combined savings can suit C.
• Do measures interact? Several measures affecting each other are hard to isolate, which points towards C or D.
• Is there baseline data? No baseline means D, or starting metering now and waiting.
• What metering exists? A well-instrumented Niagara station makes B far cheaper than it would otherwise be.
• Who has to accept the numbers? A performance contract, a utility incentive programme or an investor may specify the option and the level of rigour.

What your Niagara station needs to collect

Whatever the option, the quality of the result depends on the data. For M&V built on building automation data, that means:

• Consistent intervals — typically fifteen-minute or hourly histories for meters and key equipment.
• Raw data kept, not just rollups. Averages can be recalculated from raw histories; raw histories cannot be recovered from averages.
• Validation for gaps, spikes, meter rollovers and stuck sensors, with every correction logged.
• Independent variables recorded alongside energy: outdoor temperature from a reliable source, occupancy or schedules, production where relevant.
• Tagging, so every point’s meaning is explicit and models can be rebuilt by someone else.
• A change log of everything that might need a non-routine adjustment.

Checking the baseline model

A baseline model should be tested before anyone relies on it. The usual statistics are CV(RMSE), which measures how closely the model tracks actual use, and NMBE, which shows whether it is biased high or low. ASHRAE Guideline 14 publishes acceptance thresholds commonly used for both.

Just as important is reproducibility: the data, the model and every adjustment should be documented well enough that a third-party verifier reaches the same answer. That is what makes a saving auditable.

Our energy analytics service builds baselines this way, directly from the histories your Niagara station already collects.