Intelligent energy management systems (IEMS)

Take a 55 kW compressor that runs unloaded at about 15 kW for 60 hours a week outside production. That is roughly 47 MWh a year. The supply meter hides it inside the site total. A meter on the compressor circuit will, if someone compares its power with the production schedule. Someone with authority must then change the controls. After the change, the next month of data must show that the night load has gone.

An intelligent energy management system (IEMS) is the equipment and software that makes that sequence routine. This guide covers what to specify so that each step can be tested before you accept the system.

What is an intelligent energy management system (IEMS)?

An IEMS connects energy measurements with analysis and an operational response. The response can be an operator decision or commissioned automatic control. Products sold under the name vary widely, and some are dashboards only.

Ask which of these four functions a product includes, and keep them distinct when you compare proposals:

A threshold alert is useful without automatic control. A building management system (BMS) often already owns HVAC schedules and plant sequences. Agree where energy analysis informs the BMS and who keeps authority to change it.

Decision and measurement boundary

Write the first operational question before you select meters or software. “Reduce energy use” is too broad to commission. “Find why the compressor runs outside production hours” names a load, an operating condition and a person who can investigate it.

Choose a measurement boundary that matches the decision. A supply meter explains purchased electricity. A compressor meter explains that compressor. Photovoltaic generation or a battery behind the supply meter changes what the supply meter means. If a site consumes 150 MWh a year of its own PV output, grid imports fall by 150 MWh while consumption stays the same. Meter generation, storage, import and export separately so that the balance closes.

For plant work, the equipment measurement-boundary guide explains why energy inputs and useful outputs must describe the same system. The compressed-air monitoring guide applies that to electrical power, air delivery and operating conditions.

Specify accuracy for the whole installed chain, not the meter alone. A CT-operated meter is normally type-tested to IEC 62053-21 or IEC 62053-22 without its current transformers (CTs). The CTs add their own error, and that error grows at low load. IEC 61869-2 Table 201 allows a class 1 CT a ratio error of ±1% at rated current, ±1.5% at 20% and ±3% at 5%. A class 0.5 CT is allowed ±0.5%, ±0.75% and ±1.5% at the same points.

Now put a 400 A class 1 CT on a compressor circuit that draws 30 A when unloaded. The CT operates at 7.5% of its rating, where the permitted error is between 1.5% and 3%. The project expects to save 3% on that circuit. The measurement error is as large as the claimed saving. Size the CT for the operating current, not the breaker. The CT selection guide covers ratio, aperture and phase association.

Site and portfolio architecture

An IEMS can combine existing meters, new sensors, a local Gateway, site software and a remote platform. The proposal must state where each function runs: acquisition, history, analysis, rules and user access. It must also state which network each function needs.

A cloud platform is optional. A site can collect measurements, keep history, show dashboards and run configured rules locally. A remote platform adds portfolio comparison or specialist workflows. Specify which functions must continue during an internet outage, then test them on the proposed system.

For an existing BMS or meter network, get the actual point list and the register map for the exact meter model and firmware. “Supports Modbus” tells you nothing about which registers exist. The common integration faults are predictable:

• A 32-bit value split across two Modbus registers is read in the wrong word order. The result is either nonsense or a plausible number that is wrong.
• A scaling register or configured CT ratio is ignored, so values are out by a factor of 10, 100 or the CT ratio.
• A BACnet point is trended on change of value (COV). Readings then arrive at irregular times and only when the value moves by the COV increment.
• A pulse output is commissioned with the wrong pulse weight, for example 1 kWh per pulse instead of 10 kWh.

The Modbus RS-485 commissioning guide, M-Bus guide and pulse meter guide cover the checks for each interface. For new wireless measurements, survey the site at the meter locations and the reporting interval you need.

Ask each supplier to document these items and to show a test for each one:

A “connected” status proves only that the link is up. Get separate evidence that the site collected the reading, that it was sent, and that the receiving platform stored it with the right value, unit and timestamp. Connecting Edge to your platform covers how Edge sends data out and which platforms EpiSensor maintains a flow for.

Data quality

A dashboard must distinguish a valid zero, a missing reading, an old reading and an estimated value. Set an expected reporting interval and a freshness limit for each point class. Keep estimates marked as estimates, and keep the original observations when you correct or aggregate data.

Define completeness as valid observations divided by expected observations, for a stated point and period. A 15-minute point has 96 expected intervals on a normal day. If 91 are valid, completeness is 94.8%. On the day clocks go forward the day has 92 intervals, and on the day they go back it has 100. A system that assumes 96 reports the wrong completeness twice a year. It can also merge or drop an hour of data. Report completeness per critical point as well as across the estate, so one failed meter does not disappear inside a good portfolio average.

These are reasonable starting values:

Adjust these values for the application and record the reason. Agree how planned shutdowns affect the completeness denominator.

The brief must also define how the system handles these conditions:

• Source time against arrival time, clock synchronisation, time zones and interval boundaries.
• Units, multipliers, import and export direction, and cumulative counters against interval values.
• Gaps, duplicate arrivals, late backfill, meter resets and meter replacement.
• Counter rollover. A 16-bit counter wraps at 65,535 and a 32-bit counter at 4,294,967,295. A system that subtracts the readings without handling the wrap shows a large negative interval, or a large spike if it takes the absolute value.
• Aggregation. Adding cumulative readings does not give interval consumption. Averaging power over an hour hides a 15-minute peak.
• A reversed CT. It shows negative import on a load that cannot export. Alert on any negative value on an import-only point.

Use a commissioning sample that includes a known gap, a late reading and a meter reset. Ask the supplier to show the original records and the resulting chart or calculation.

Baseline and savings

A saving is the difference between measured consumption and what consumption would have been without the change. A lower bill or a before-and-after chart does not isolate the change. EVO's IPMVP principles require consistent boundaries between the baseline and reporting periods, with adjustments for the conditions that differ.

Consider a plant with this record:

The bill fell by 40 MWh, or 9.5%. The kWh per unit rose by 8.6%. Neither figure tells you what the new controls did. Suppose regression on the baseline data gives a fixed load of 150 MWh a year plus 0.225 kWh per unit. The adjusted baseline for the reporting year is 150 MWh + 1.0 million × 0.225 kWh = 375 MWh. The plant used 380 MWh, which is 5 MWh more than expected. The fall in the bill came from lower production, and the simple ratio exaggerates the loss because the fixed load is spread over fewer units.

ISO 50006 gives guidance on energy performance indicators (EnPIs) and energy baselines. It covers the choice of relevant variables such as production and heating or cooling degree-days, and when to revise a baseline. For degree-days, state the base temperature, because the result changes with it. Keep the original baseline and a versioned record of every adjustment. Record non-routine changes, such as a new production line or a change of floor area, separately from routine adjustments.

ASHRAE Guideline 14 judges a regression baseline by its coefficient of variation of the root mean square error, CV(RMSE), and by its net bias. For example, the prescriptive whole-building path in the 2002 edition requires a full 12-month baseline. It also requires a CV(RMSE) of no more than 20% for energy when less than 12 months of reporting data is used, and expected savings above 10% of the metered use. The limits changed in later editions, so name the edition in the contract.

ISO 50015 gives general guidance for measurement and verification of organisational energy performance. The U.S. Department of Energy's summary of the IPMVP options distinguishes four methods:

• Option A isolates the retrofit, measures the key parameter and estimates the rest. For a lighting retrofit, the key parameter is often the fixture power, and the operating hours are estimated.
• Option B isolates the retrofit and measures all parameters. This usually means continuous metering of the isolated system.
• Option C analyses the whole facility from utility or supply meters. The saving must be large compared with the model error.
• Option D uses calibrated simulation.

Before a performance-based procurement, the M&V practitioner and the supplier agree the boundary, option, baseline and reporting periods, explanatory variables, missing-data treatment, uncertainty and reviewer. The report must be reproducible from retained inputs. Consult EVO's current protocol library for the project method.

Keep energy, cost and emissions results separate. A cost result needs the tariff and its time bands. An emissions result needs an accounting boundary, a factor source and a factor date. Every percentage saving states its baseline, scope, period and method.

Demand management and automatic control

Energy is billed in kWh. Demand or capacity is billed in kW or kVA, usually from the highest interval average in the billing period, commonly over 15 or 30 minutes. Read the exact quantity, interval and rule from the site's tariff and connection agreement.

The ESB Networks statement of charges is a clear example. Maximum-demand customers pay an annual capacity charge per kVA of maximum import capacity (MIC). If peak demand in kVA exceeds the MIC, a quarter-hour (QH) metered connection pays a surcharge of five times the capacity charge rate on each excess kVA for that billing period. A connection without QH metering pays six times. For a connection with no export capacity, a low power factor surcharge applies to the kVArh above one third of metered kWh. That threshold is a power factor of about 0.95.

Because the MIC is in kVA, power factor changes the bill. Take a site with a 450 kVA MIC and a 400 kW peak:

The same site can also stay under its MIC by moving one load. If a 60 kW chiller starts at 07:45, in the same interval as the production lines ramp up, it adds to the peak. Delaying its start by 30 minutes moves it out of that interval. The kWh do not change. Use the power factor correction calculator and the interval demand calculator to work through your own figures, then check the result against the supplier's settlement data. The DOE rate guide explains why load shifting affects energy and demand charges differently. Its examples are U.S. tariffs.

Demand-response participation adds programme-specific baselines, metering, availability and verification requirements. Assess these with the service provider through the demand-response project journey.

For automatic control, write a sequence that names the trigger, the permitted action, the maximum duration, the equipment feedback and the recovery state. The sequence must also cover stale inputs, lost communications, a Gateway restart, a failed command and manual override. Equipment interlocks and approved safety controls stay authoritative.

Test the physical result of a rule, not only its execution log. A command sent to a controller does not prove that the equipment responded: an acknowledged command and a measured fall in load are two different observations. Make one system responsible for each controlled function, so that a BMS schedule and an energy rule cannot compete without anyone noticing.

If a proposal includes AI or forecasting, ask for a comparison against a simple method, such as the same day last week, on data withheld from model development. Agree the error metric, for example CV(RMSE) or mean absolute percentage error, and the margin by which the model must beat the simple method. Agree a limit for false alerts per site per week. An alert that takes an engineer 30 minutes to investigate and fires 10 times a week costs 5 hours of staff time every week.

Governance, commissioning and acceptance

Tie each role to a decision it signs. The installer signs point acceptance. The site operator approves control rules and overrides. The energy owner approves baseline revisions. The M&V reviewer signs the savings report. The IT owner controls user access, software updates and backup restoration. Include access removal, incident escalation and configuration history in the handover.

ISO 50001 applies to the organisation's energy management system, including responsibilities, EnPIs and continual improvement. Metering and software supply evidence for it. Installing a product does not give conformity or certification.

Turn the specification into signed acceptance criteria before the pilot begins:

1. Point acceptance: reconcile the installed point list with physical assets. Check units, direction, scaling and timestamps against a reference instrument or a fiscal meter, within an agreed tolerance. Qualified electricians fit CTs and voltage connections.
2. Data acceptance: meet the agreed freshness and completeness limits over a representative operating period that includes a weekend. Inspect gaps, duplicates and counter events at the destination.
3. Outage acceptance: interrupt the network for a set time. Show the agreed local functions working, then confirm that every retained interval arrives once and within the specified recovery time.
4. Control acceptance, where included: witness authorised actions, actual feedback, overrides and failure recovery against the approved sequence.
5. Analysis acceptance: reproduce a selected report from exported inputs, with its baseline version, adjustments and excluded data.
6. Handover acceptance: deliver configurations, point mappings, access ownership, operator training and a tested recovery procedure.

Keep technical acceptance of the monitoring system separate from any later finding that an efficiency measure delivered its expected saving.

Compare total project cost over the service life: surveys, installation, communications, licences, integration, verification, maintenance and support. Include the staff time spent investigating findings, which is often larger than the licence fee.

IEMS with EpiSensor

EpiSensor Edge runs on the ZGW-20 Gateway at the site. It gives local device management, dashboards and historical data analysis. Edge records readings on the Gateway, so local history and dashboards stay available when the WAN is down. How far back that history goes depends on the recording settings and retention. External notifications and onward delivery to a platform need the network.

Rules run on the Gateway through Edge's Automation extension, when it is installed with the device integrations the rule uses. When the extension evaluates a rule, it ignores any reading older than its stale-data threshold, which is five minutes by default. A point that has stopped reporting therefore does not trigger an action.

ZEM electricity monitors measure to Class 0.5S under IEC 62053-22, with the meter and its current sensors calibrated together before shipment. Readings go over the Zigbee mesh to the Gateway. Existing meters connect through the ZMB Modbus interface, the ZHM M-Bus interface or the ZPC pulse counter. Use Edge as the site system, or send data to a chosen platform where portfolio analysis or specialist M&V needs one.

The CoolPlanet manufacturing case study shows circuit data used to find hot oil boiler trips at weekends and chiller and compressor loads outside production hours. CoolPlanet combined those measurements with its own software and engineering work at one client, so its savings figures belong to that site.

For an estate, the Royal Mail case study describes a pilot at the Cardiff Mail Centre. It sent data to the estate's existing energy management platform and led to a rollout across 40 mail centres. Use the multi-site application guide to define point naming, site autonomy and rollout checks for your own sites.

To start an energy-management project, send the site question, the existing meter list, the reporting interval you need and the name of the person who will act on the result.