Energy Performance & ISO 50001

Q: We are not ISO 50001 certified. What does the module tell us?

The same thing, just to management instead of an auditor: how much of the change in consumption came from production, and how much from efficiency. If you start the certification process, the EnB and EnPI infrastructure is already in place; if you never do, the monthly performance conversation still moves from guesswork to measurement.

Q: Why freeze the baseline? Isn't updating it as the model improves more accurate?

The model improves, as versions. If the baseline is reworked every month, the 'savings' figure changes retroactively every month too, and that destroys trust in front of auditors and boards alike. This is precisely ISO 50001's EnB logic: the reference is fixed; revisions are dated and justified.

Consumption fell last month. Did you save energy, or did production fall? Every energy report that cannot separate the two tells management the wrong story.

Get in touch

Deviation

How far from expected this month?

Raw consumption lies: when production falls, consumption falls too, and that is not a saving. Expected consumption is built from production and weather; the deviation is written as a percentage.

Baseline: a frozen reference period (EnB)
Months over the threshold are flagged
Ready for the ISO 50001 EnPI report

The baseline is frozen from a reference period (EnB) and does not change afterwards; every month’s expected consumption comes from the same model. Deviation is kept as both a percentage and kWh; this table is presented as evidence in the ISO 50001 internal audit.

Variables: production, degree-days, operating hours
Threshold ±5% by default, set per site
Source-based model: electricity and gas separately

Consumption Forecast

Forecast modelDeviations

Each point is a period. The line is the learned relationship; the band is the forecast interval.

Regression

The relationship is visible

Each point is a period: production against consumption. The line is the relationship the model learned; next month’s expected consumption is read from it. A weak fit (low R²) is written down, not hidden.

The model is built per energy source; alongside production, variables such as degree-days and operating hours can be added. If R² is low the panel warns “weak relationship” and widens the forecast band; nobody meets false precision.

Hovering a point opens that period’s production, consumption and expected value; months above the line consumed more than expected. The band is the forecast’s confidence interval: a point inside it is noise, one outside is an event to examine.

Each point is a month or a week
The forecast band widens with a low R²
The reference period is frozen

Regression Analysis

reference period 2023 · 14 observations · R² = 0.14

y = 82,628 + 7.9 × Production(t). The scatter of the points says what the R² says, visibly: at this site, tonnage alone does not explain consumption.

CUSUM

The cumulative result on one screen

The truth of the accumulation is separated from the noise of a single month. If the curve bends down, the saving is holding; if it turns up, something changed. This curve is the evidence for an efficiency project.

Each month’s deviation is added to the previous one; the slope of the curve is the speed of saving. The month an efficiency project was applied is marked, and the slope after it is compared with the slope before; the ESOS evidence and the savings verification report come from the same curve.

The curve updates automatically every month; in the pounds view the deviation is multiplied by that month’s weighted price. Several sources (electricity, gas) sit as separate curves on the same screen; the total CUSUM tells the story of the whole site.

Project start marked on the curve
Cumulative kWh and pounds together
Monthly report by e-mail

CUSUM · Cumulative Deviation

Σ(actual − expected) · downward = saving

Savings accumulated over two months: 5,890 kWh. The slope shows speed, steepening means the improvement is biting, flattening means its effect is spent.

What we measure, and from where

Measurement provenance

Consumption: kWh; Main meter and area analysers · hourly
Production: tonnes · units; ERP/MES link or periodic entry
Degree-days: °C·day; Hourly weather · by site location
Baseline: equation; Regression · published with R² and observation count
Deviation: %; Derived · actual − expected
CUSUM: kWh; Derived · cumulative deviation
EnPI: kWh/driver; Defined per site · tied to the EnB version

Frequently asked

We are not ISO 50001 certified. What does the module tell us?

The same thing, just to management instead of an auditor: how much of the change in consumption came from production, and how much from efficiency. If you start the certification process, the EnB and EnPI infrastructure is already in place; if you never do, the monthly performance conversation still moves from guesswork to measurement.

What do you do when R² comes out low?

First, we say so, we do not hide it. Then we hunt the missing variable: degree-days at a cooling-heavy site, product mix on a multi-product line, shift count in shift operations. When a variable is added and the model strengthens, a new baseline version is opened; the old version and every report produced with it stay in the archive untouched. We do not rewrite history.

Why freeze the baseline? Isn't updating it as the model improves more accurate?

The model improves, as versions. If the baseline is reworked every month, the 'savings' figure changes retroactively every month too, and that destroys trust in front of auditors and boards alike. This is precisely ISO 50001's EnB logic: the reference is fixed; revisions are dated and justified.

We don't want to hand over production data. Does the module work without it?

In a limited way. Without driver data there is no normalisation; what remains is raw-consumption comparison, the very trap in this page's first paragraph. For sites that will not share absolute tonnes, we run indexed: a production index relative to the reference period does the same job, and no commercial data leaves the site.

Are the numbers on screen real?

Yes, the regression and deviation analysis are the real 2023-referenced data of our own pilot site; that is exactly why an unflattering R² of 0.14 is left on screen. Customer site data is never used on this website.