Industrial Energy Optimisation: Baselines, EnPIs and ISO 50001

Industrial energy optimisation is the discipline of improving a plant's energy performance with measured data: a baseline that models expected consumption, indicators that separate management effects from calendar effects, and the ISO 50001 cycle that keeps the whole thing alive. This guide walks the full loop through one worked example, a 4.8 GWh/year packaging plant, from baseline to verified savings.

What industrial energy optimisation is and why it matters

Industrial energy optimisation is the discipline of continuously improving a plant's energy performance on the strength of measured data, not catalogues of tricks. ISO terminology distinguishes three dimensions of that performance: efficiency (how much energy I need per unit of output), use (what I spend it on) and consumption (how much I spend in total). Optimising means acting on all three with method.

The economic incentive is direct. In intensive sectors such as paper, ceramics or chemicals, energy can account for between 10% and 40% of operating cost. Even in medium-intensity industries such as food and beverage, a plant turning over 20 million euros can be paying more than 600,000 euros a year for electricity and gas.

Add to this the regulatory pressure. In Spain, Royal Decree 56/2016, the national transposition of the European Energy Efficiency Directive, obliges large companies to carry out an energy audit every four years; the decree itself recognises a certified ISO 50001 energy management system as an alternative. The revised directive, (EU) 2023/1791, replaces the company-size criterion with annual consumption.

The idea running through this article fits in one sentence: you cannot optimise what you do not measure against a reference. That reference is called the energy baseline, and building it well is half the work. We will see it through a complete example, a packaging plant consuming 4.8 GWh of electricity a year, which will stay with us to the end.

The energy baseline: the reference everything is measured against

ISO 50001 defines the energy baseline (EnB) as the quantitative reference characterising energy performance over a given period. The definition sounds harmless, but it hides the most important methodological decision in the whole system: the baseline is not a figure, it is a model.

The usual temptation is to compare absolute consumption. «Last year we used 4.9 GWh and this year 4.8: we have improved by 2%». That reasoning fails because consumption depends on variables nobody controls from the boiler room. If production was 5% lower this year, consumption should have fallen by more than that 2%: performance actually got worse. If the winter was mild, the boiler worked less without anyone touching anything. Comparing kWh against kWh mixes the effect of management with those of production, weather and product mix.

The solution is to express the baseline as a model of expected consumption as a function of the relevant variables. With it, the right question stops being «did I use less than last year?» and becomes «did I use less than the model predicted for this period's actual conditions?». That difference separates accounting from energy management.

Nor is the baseline eternal: it must be revised after structural changes (a new production line, the replacement of a furnace, a change of product). Keeping an obsolete baseline turns every subsequent indicator into noise.

Relevant variables: production, degree days and shifts

Choosing the model's variables is an exercise in plant physics before it is one in statistics. A candidate variable must meet two conditions: a plausible physical mechanism connecting it to consumption, and reliable data for it at the same frequency as the energy data.

Production is the main driver in almost any process. It can be measured in units, tonnes or litres; what matters is that it represents useful work. In plants with products of very different energy intensity it pays to disaggregate by product family.

Degree days quantify the thermal load of the climate. Heating degree days add up, for each day of the month, the difference between a base temperature (usually 15 or 18 °C) and the mean outdoor temperature when the latter falls below it; cooling degree days do the mirror image. They are the natural variable for HVAC, air-cooled refrigeration and processes sensitive to outdoor conditions. In Spain they can be calculated from the daily data published by AEMET, the national meteorological agency.

The third group are the operating-regime variables: shifts, running hours, working days, occupancy. A plant moving from two to three shifts changes its consumption-production relationship even if nothing else changes.

The practical rule: start with one or two variables, verify that the model works and add complexity only if the residuals demand it. A model with five variables and twelve monthly data points is not a model, it is a forced fit.

Building the baseline by regression: step by step

On to the example. Our packaging plant consumes 4,800,000 kWh a year and produces 10,000 tonnes. These are its twelve months:

The most useful model to start with is the linear regression of monthly consumption E against monthly production P:

E = a·P + b

a = Σ(Pi - P̄)(Ei - Ē) / Σ(Pi - P̄)²
b = Ē - a·P̄

With P̄ = 833.3 t and Ē = 400,000 kWh, the calculation gives a = 330 kWh/t and b = 125,000 kWh/month (rounding 329.97 and 125,027). The baseline becomes:

Expected consumption (kWh/month) = 330 × Production (t) + 125,000

Both coefficients have a direct physical reading. The intercept, b = 125,000 kWh/month, is the fixed load: what the plant consumes even when producing nothing (compressors idling, refrigeration that never switches off, lighting, equipment on standby). That is 1.5 GWh a year, 31% of the total, and it is the first figure that should make any plant manager uncomfortable. The slope, a = 330 kWh/t, is the marginal consumption of producing one additional tonne.

Before using the model it must be validated. The R² coefficient measures what fraction of the variability in consumption is explained by production; here it comes out at 0.97. As a rule of thumb, below 0.75 a variable is missing or data errors abound. ASHRAE Guideline 14 and the IPMVP prefer the CV-RMSE (the model's standard error divided by mean consumption): 8,200 kWh over 400,000, or 2.1%, well below the usual 15% limit for monthly data. It also pays to look at the residuals: if the summer months always sit above the line, the climate is asking to enter the model.

When thermal load matters, the model expands into multivariable regression:

E = a·P + c·DD + b

where DD are the month's degree days and c is read in kWh per degree day. The mechanics are the same (least squares) and any spreadsheet solves it.

Three mistakes ruin baselines: fitting with fewer than ten or twelve points; extrapolating outside the observed range (our model is valid between 520 and 980 t/month, nobody knows what the plant does at 300); and adding spurious variables that raise the R² with no physics behind them, such as the month number or the price of energy.

Energy performance indicators (EnPIs)

An energy performance indicator (EnPI) is the quantity chosen to track performance. ISO 50006 is the reference guide, and its central message is that there is no universal indicator: each one answers a different question.

The most widespread is specific consumption:

SC = energy consumed / units produced   (kWh/t, kWh/unit...)

Our plant has an annual SC of 4,800,000 / 10,000 = 480 kWh/t. It is intuitive and useful for comparing similar plants. But fixed load distorts it at low output, and the example gives it away: in August the plant produced 520 t at an SC of 560 kWh/t; in October, 960 t at 459 kWh/t. Was it 22% less efficient in August? No: the 125,000 fixed kWh were spread across fewer tonnes. The model itself explains it: expected SC = 330 + 125,000/P, a curve that shoots up as P falls. Judging the August crew by their specific consumption is punishing them for the holiday calendar.

Energy intensity carries the idea into economic terms: energy divided by value added or turnover (kWh/€). It is the indicator of sector statistics, but it mixes efficiency with prices and product mix: it serves the board, not the morning shift.

The most honest indicator is the model-based one: compare actual consumption with expected consumption according to the baseline and accumulate the differences. That cumulative sum is called CUSUM: while the plant behaves like its baseline, it oscillates around zero; when something improves or worsens in a sustained way, the curve changes slope and makes it visible long before any annual average would.

In practice a hierarchy works best: at plant level, deviation against baseline and monthly CUSUM; at system level (compressed air, refrigeration), specific consumption with weekly submetering; at critical-equipment level, powers and efficiencies as the available instrumentation allows. Each level must answer to someone specific, or nobody looks at it.

ISO 50001: structure and the PDCA cycle applied to energy

ISO 50001 is not a catalogue of measures but the specification of a management system: the structure that guarantees energy is planned, controlled and improved continuously. Since 2018 it has shared the high-level structure (HLS) with ISO 9001 and ISO 14001, which allows it to be integrated into the systems most plants already run. Anyone managing quality with PDCA already knows how to manage energy; the object changes, not the method. It is the framework on which Captia builds its energy efficiency projects in plant. Here we stick to the cycle in application; the standard clause by clause, phased implementation and the certification process have their own guide: industrial energy efficiency with ISO 50001.

Plan. The phase with the most technical content: the energy review (next section), the baseline, the EnPIs, and from there quantified objectives with owners, deadlines and resources. «Cut compressed air 5% against baseline before December» belongs to this phase; «raise energy awareness» is not an objective, it is an intention.

Do. Run daily operations with energy in mind: training for the staff who influence the significant uses, documented operational controls (setpoints, start-up schedules, temperatures), and two requirements that often go unnoticed: incorporating energy performance into the design of new facilities and into procurement. Choosing the cheap motor over the efficient one is a decision the plant pays for over twenty years.

Check. Tracking the EnPIs against the baseline, verifying savings, internal audit. It is the phase that degenerates fastest without automated data: filling in spreadsheets by hand the day before the audit is not monitoring.

Act. The management review: analyse results, correct, update the baseline and objectives if there were structural changes, and allocate resources to the next cycle.

Certification brings discipline, continuity through staff turnover, exemption from the four-yearly RD 56/2016 audit (provided the system includes that audit) and a credible signal to customers. What certification does not do is save energy on its own: there are certified plants with mediocre performance and uncertified plants with excellent management. The certificate attests to the system; the kWh are saved by operations.

The energy review: identifying significant energy uses (SEUs)

The energy review is the initial diagnosis: how much energy comes in, where it goes and where to concentrate effort. It has three steps.

First, the inventory by source: electricity, gas, diesel, with their annual consumption drawn from invoices and meters. Second, disaggregation by system: splitting each source across the main consumers through submetering where it exists and, where it does not, through estimates based on installed power and operating hours (documenting the assumption).

In our plant, the electrical breakdown looks like this:

Third, the Pareto analysis: rank by consumption and confirm what almost always holds, that three or four systems concentrate more than 85% of the total. Onto that ranking the significance criteria are applied: a use is declared significant (an SEU) when it weighs heavily in consumption, when it has considerable improvement potential, or both. Process is the biggest consumer here, but its margin is limited by the technology of the lines; compressed air, with 18% of consumption and its chronic inefficiencies, is the perfect candidate.

Declaring an SEU is not sticking on a label: for each one you must identify the variables affecting its consumption, the people who influence it and the associated operational controls. For compressed air: variables, production and setpoint pressure; people, maintenance and the operators of the blow guns; controls, the documented setpoint and the monthly leak round.

Typical optimisation measures on the shop floor

With the baseline built and the SEUs identified, the measures arrive. The figures that follow come from public technical literature (the US DOE, guides from IDAE, the Spanish energy agency) and are given as ranges with validity conditions, because a percentage without context is advertising, not engineering. The prioritisation is constant: near-zero-cost operational measures first (setpoints, schedules, leaks), then investments, ranked by payback.

Compressed air: the prime suspect

Compressed air is the most expensive form of energy in a plant, and it is so by thermodynamics: of the electricity entering the compressor, only in the order of 10-15% ends up as useful work at the point of use; the rest is lost in compression heat, pressure drops and leaks.

Leaks are the star measure. In plants without a detection programme, DOE literature puts losses at between 20% and 30% of the air generated; a well-maintained system stays below 10%. The cost is quickly calculated: a 3 mm orifice at 7 bar lets out about 11 l/s, forcing the compressor to supply about 4 kW continuously. Over 6,000 annual hours that is about 24,000 kWh, around 3,400 euros a year at 0.14 €/kWh. For a hole the size of a lentil. Detection is done with ultrasonics during shutdowns; the repair is usually fittings and labour.

Two measures complete the basic trio. Lowering the setpoint pressure: each bar less saves around 7% of compression energy, and many networks run at 7.5 bar because a single old machine once demanded it. And recovering the compressor's heat: between 70% and 90% of the electricity consumed is recoverable as hot air or hot water, if there is demand nearby.

Electric motors and variable speed drives

Motor-driven systems (pumps, fans, compressors, conveyors) account for around two thirds of industrial electricity consumption, according to the International Energy Agency. It is the largest item and the most regulated: the European ecodesign regulation 2019/1781 requires minimum IE3 efficiency for most new motors since 2021, and IE4 between 75 and 200 kW since 2023. Between IE2 and IE4 the difference looks small (two or three points of efficiency), but a motor runs so many hours that the price premium almost always pays back.

The physics to know are the affinity laws of centrifugal machines: flow is proportional to speed, pressure to its square and power to its cube. Reducing a pump's speed by 20% leaves power at 51% (0.8³ = 0.512). That cube is the reason variable speed drives exist.

From it comes the criterion for when a drive saves energy: when demand is variable and is currently regulated by destroying energy, with throttling valves, dampers or bypasses. There, savings of 20-50% of the motor's consumption are plausible. With a constant, well-matched load, the drive saves nothing and adds its own losses (2-4%). It is not an efficiency measure in itself; it is a better way to regulate.

The third front is oversizing: a motor chosen «with margin» works far from its optimum point and an oversized pump forces permanent throttling. Reviewing the actual operating point against the design point is one of the cheapest audits there is.

Lighting and waste heat

Lighting is the mature measure par excellence. Moving from fluorescent or discharge lamps to LED with presence detection and daylight dimming saves between 50% and 70% of lighting consumption, with paybacks of one to three years in halls with long operating hours. In our plant, lighting is 336,000 kWh/year; a 60% saving is worth about 28,000 euros annually. It will not transform the P&L (it is 7% of consumption), but it is fast and visible: a good first measure, because it builds credibility for the ones that follow.

Waste heat is the opposite case: large potential and more engineering. The usual sources are boiler and furnace flue gases, steam condensate, compressor heat and cooling circuits. The technologies, from least to most complex: stack economisers, air-to-air or water-to-water heat exchangers, and industrial heat pumps capable of lifting heat from 30-40 °C up to a useful 70-90 °C.

Feasibility is decided with three questions: temperature (is the heat at a sufficient level for the demand or does it need pumping up?), distance (are source and demand close enough that the pipework does not eat the project?) and simultaneity (does the demand exist when the source produces?). With all three favourable, heat recovery is usually among the most profitable measures in plants with thermal demand.

Measurement and verification of savings: the IPMVP approach

Here a conceptual discomfort appears: savings cannot be measured. No meter reads saved kWh, because the saving is the difference between what the plant consumes and what it would have consumed without the improvement, and that second term is a counterfactual: it does not exist. It can only be estimated with a model. That is what we built the baseline for:

Saving = expected consumption (baseline adjusted to the period) - actual consumption

The IPMVP protocol (International Performance Measurement and Verification Protocol, maintained by EVO) is the reference standard, with four options:

In all of them, the baseline is adjusted to the verification period: routine adjustments are those the model makes when the period's actual production or degree days are fed in; non-routine adjustments respond to structural changes (a new line, an extra shift) that force a correction to the model.

Let us close the example. The plant executes the obvious measure on its SEU: an ultrasonic leak detection and repair campaign, plus lowering the setpoint from 7.5 to 7.0 bar. Total cost, about 9,000 euros. The following six months, against the baseline:

Actual consumption sits below expected every single month, by an average of 11,000 kWh. Annualised: about 132,000 kWh, 2.75% of the plant's consumption and 15% of the compressed air system, consistent with the leak literature. In euros, 132,000 × 0.14 = 18,480 a year. The investment pays back in under six months.

One nuance remains, the one that separates a serious report from a sales one: uncertainty. The baseline's standard error is 8,200 kWh/month; over a half-year, the 95% confidence band on the cumulative saving is around ±40,000 kWh. The saving is real (all six months fall on the same side), but its exact value still has a wide range. Extending the tracking narrows it; submetering the compressor (moving to option B) would narrow it far more. Never report a saving without its confidence band.

Continuous monitoring: from the annual audit to daily data

The classic energy audit is a still photograph: an external team measures for a few days, delivers a report and disappears for four years. The photograph has value, but energy is a motion picture. The optimised setpoint gets nudged back up in the first production squeeze; the repaired leak has new siblings every month. Without continuous tracking, savings decay and nobody notices until the bill.

The alternative is to treat energy the way quality is treated: with continuous data and statistical control. The minimum architecture has four layers: meters on each SEU (the utility's main meter is not enough), acquisition every few minutes, historical storage and an analysis layer comparing actual against expected consumption. It is exactly the approach of the connected energy management Captia deploys on the shop floor. The analysis layer of that architecture is the Energy Optimization module in Captia.ai, where the data it consumes, what it returns and when it does not apply are detailed.

In that layer, the baseline stops being a spreadsheet someone opens once a month and becomes a live algorithm: every day, the system calculates expected consumption from the actual production and weather, subtracts it from the measured figure and accumulates the difference in a CUSUM per SEU. A sustained drift triggers an alert while correcting it costs a call to maintenance, not after it has spent three months paying itself into the bill. It is the same mechanics as the compressed air verification, executed continuously.

This is not an extra alien to the standard: the monitoring, measurement and analysis chapter of ISO 50001 asks for exactly this function. The maturity rule is simple: if the measurement plan fits in one person with a spreadsheet, the management system will last as long as that person does.

The complete worked case: the 4.8 GWh/year plant

Let us recap the whole journey, because the order of the steps is the method:

1. Data. Twelve months of consumption and production: 4,800,000 kWh and 10,000 t. Invoices and production reports, nothing exotic.
2. Baseline. Monthly regression: E = 330·P + 125,000, with R² = 0.97. Fixed load of 125,000 kWh/month (31% of the total) and 330 kWh per marginal tonne.
3. Indicators. Annual SC of 480 kWh/t, but it swings between 459 and 560 depending on the month because of the fixed load: the management indicator is the deviation against the model, not the ratio.
4. Energy review. Pareto by system: process 55%, compressed air 18%, refrigeration 15%, lighting 7%. Compressed air is declared an SEU on weight and potential.
5. Measure. Leak campaign and setpoint lowered to 7.0 bar: 9,000 euros.
6. Verification (IPMVP option C). Six months against baseline: 66,000 kWh below expected, 132,000 kWh annualised, 18,480 euros a year, payback in under six months.
7. Tracking. The baseline moves into the monitoring system as a daily algorithm with CUSUM and alerts; management reviews the cycle once a year.

No step demands mathematics beyond a regression: it demands method and persistence. Want to translate these percentages to your plant? Try your own numbers:

Frequently asked questions

References and recommended reading

Standards (paid, through AENOR/ISO):

• ISO 50001:2018. Energy management systems. Requirements with guidance for use.
• ISO 50006:2023. Evaluating energy performance using energy baselines and energy performance indicators.
• ISO 50015:2014. Measurement and verification of energy performance.
• ISO 50002. Energy audits.

Protocols and application guides (free):

• IPMVP Core Concepts, Efficiency Valuation Organization (EVO). The reference text in measurement and verification.
• ASHRAE Guideline 14. Statistical validity criteria for models (CV-RMSE).
• Improving Compressed Air System Performance: A Sourcebook for Industry, U.S. DOE. Source of the figures on leaks, pressure and heat recovery.
• Premium Efficiency Motor Selection and Application Guide, U.S. DOE. Motors, IE classes and affinity laws.
• Technical guides from IDAE (the Spanish institute for energy diversification and saving): audits, industrial lighting and cross-cutting systems, in Spanish and freely available.
• Royal Decree 56/2016 and Directive (EU) 2023/1791, in the Spanish official gazette (BOE) and EUR-Lex.

Everything shown in this article can be solved with a spreadsheet; to go deeper into the regression models, the IPMVP itself and Guideline 14 are the best starting point.

At Captia Technology we help industrial plants build and maintain this complete cycle, from the baseline to continuous monitoring, on their own production and energy data: from energy efficiency to energy certifications. If you would like to see what it would look like at your plant, write to us and we will set it up on a real example of your consumption data.