Peak Demand Reduction for Industrial Facilities: Metering, Load Control & BESS Sizing

Peak demand reduction is an interval-control problem before it is an equipment-selection problem. Start with the utility tariff and revenue-meter data, reconstruct the quantity that creates the demand charge, identify the processes responsible for the highest chargeable intervals, and then choose the least disruptive combination of scheduling, load control, and energy storage.

A battery energy storage system (BESS) is useful only when its power conversion system can cover the instantaneous reduction and its usable energy can sustain that reduction for every relevant event. It is not the default answer. A plant may obtain the required demand limit through staggered starts, managed charging, variable-frequency drive (VFD) setpoint control, thermal-load scheduling, temporary load shedding, storage, or a coordinated combination.

Mühendislik kapsamı: The equations below support feasibility analysis and preliminary specification. The utility’s tariff and revenue meter determine the bill. Detailed design still requires the applicable electrical code, interconnection rules, protection study, fire code, equipment documentation, and approval by the authority having jurisdiction (AHJ).

1. Define the Quantity That Must Be Controlled

Energy and demand are related but not interchangeable:

• Energy, kWh: power integrated over time.
• Demand, kW: a power value determined under the tariff’s measurement method and time window.
• Apparent demand, kVA: used by some tariffs instead of, or in addition to, kW demand.
• Billing demand: the value the tariff actually uses after applying time-of-use periods, ratchets, contract demand, power-factor adjustments, or coincident-peak rules.

For a simple fixed interval of duration (T), the average grid-import demand in interval (k) is:

P_d,k = (1 / T) × ∫ P_grid(t) dt

If the interval contains energy import (E_k), the equivalent form is:

P_d,k = E_k / T

For a tariff that bills the maximum interval only within an eligible time window (W):

D_month = max(P_d,k), for every interval k inside W

This is a model, not a universal tariff definition. Utilities may use block intervals, rolling intervals, thermal demand, coincident system peaks, several time-of-use demand registers, or minimum/ratcheted billing demand. The U.S. Department of Energy (DOE) distinguishes non-coincident demand, time-of-use demand, system-peak demand, and demand ratchets in its utility-rate guidance.

Tariff inputs to extract before analyzing equipment

Tariff input	Mühendislik sorusu	Why it changes the control design
Demand quantity	Is demand billed in kW, kVA, or both?	A kVA charge may require power-factor analysis, not only real-power reduction.
Interval method	Block, rolling, thermal, or another method?	The controller must estimate the same quantity closely enough to act before the billing value is set.
Chargeable window	All hours, on-peak hours, or several periods?	A control action outside the chargeable window may have no demand-charge value.
Ratchet/look-back	Does prior demand establish a billing floor?	Lower physical peaks may not lower the bill immediately.
Coincident peak	Is the charge based on facility maximum or grid-system events?	Forecasting an external event may matter more than capping the facility maximum.
Export rule	Is reverse power permitted and how is it metered?	The BESS controller needs an export constraint and separate import/export logic.
Contract/minimum demand	Is there a demand floor or contracted capacity?	Reducing below that floor may provide no further tariff benefit.

Do not infer any of these values from a typical 15-minute example. Read the active rate schedule, riders, special conditions, and recent bills. Where possible, confirm the interpretation with the utility.

2. Reconstruct the Baseline from Interval Data

Use at least one complete operating year when seasonality, production mix, weather, or tariff seasons can change the peak. More data may be required where the tariff has a longer look-back or the facility has infrequent campaigns.

Data-quality checks

Before ranking peaks, verify:

1. Timestamp convention: local time, UTC, daylight-saving transitions, and interval-ending versus interval-starting timestamps.
2. Interval duration: do not mix 5-, 15-, 30-, and 60-minute records without resampling by energy.
3. Missing and duplicate records: flag them; do not silently fill a peak interval with an average.
4. Units and scaling: verify Wh/kWh, W/kW, current-transformer (CT) ratio, potential-transformer ratio, and multiplier.
5. Import and export sign: store them separately when possible. A net value can conceal simultaneous load and generation behavior.
6. Meter boundary: identify which feeders, generators, and storage systems are upstream or downstream of the measurement point.
7. Revenue reconciliation: compare reconstructed monthly peaks and energy totals against bills. Explain the residual before sizing controls.

An internal meter is invaluable for real-time control, but it does not replace the revenue meter. Its clock, aggregation rule, CT accuracy, sampling rate, and meter boundary can create a systematic difference. For metering architecture and accuracy concepts, see the VIOX guide to energy meters, CTs, and Modbus.

Build a peak-event table, not only a load-duration curve

A load-duration curve shows how often demand exceeds a threshold, but it removes time order. Storage sizing and process control need chronology. Create an event table containing:

• start and end time;
• maximum and average demand;
• duration above candidate caps;
• production state, batch, shift, and weather context;
• large load starts and overlapping duty cycles;
• controllable-load availability;
• on-site generation and storage state;
• preceding and following recovery time.

Group adjacent exceedances into one event. Two peaks separated by five minutes may be one storage duty cycle if the battery cannot recharge between them.

3. Separate Five Different Demand Strategies

Treating every measure as “peak shaving” hides important differences.

Strategy	Neler değişiyor	Energy consequence	Control implication
Verimlilik	Reduces power required for the same useful output	Usually reduces kWh and may reduce peak kW	Persistent measure; verify that the saving occurs during chargeable peak intervals.
Load shifting	Moves an energy-consuming task to another time	Often similar total kWh	Requires schedule flexibility and a safe off-peak window.
Yük atma	Temporarily stops or curtails a load	May reduce output or require later recovery	Needs priority, maximum interruption time, and restart logic.
Peak shaving	Holds grid import below a cap	May shift energy or add conversion losses	Requires closed-loop measurement and an actuator such as controllable load, storage, or generation.
Demand response	Changes load in response to a utility or market event	Contract-specific	Requires event communications, baseline rules, performance verification, and operational commitment.

Oak Ridge National Laboratory distinguishes temporary load shedding from moving consumption through load shifting in its electric utility bill guide. DOE’s industrial demand-response guidance similarly emphasizes understanding the facility’s load profile before committing production loads to an event response.

Sequence low-complexity controls first

A technically rational order is:

1. eliminate metering and tariff-model errors;
2. remove avoidable simultaneous starts and recovery peaks;
3. reschedule deferrable loads;
4. modulate flexible loads within process limits;
5. shed approved loads during exceptional events;
6. use BESS or another dispatchable resource for the residual that cannot be managed operationally.

This is not a mandatory economic ranking. It prevents a storage system from being sized to cover peaks that a scheduling correction could remove.

4. Set the Demand Cap from Operations, Not from a Percentage Target

Let (D_{base,m}) be the reconstructed billing demand for month (m). A candidate cap (P_{cap}) is feasible only if the control system can cover every relevant exceedance without violating process, equipment, tariff, or interconnection constraints.

Define the uncontrolled net import profile as:

P_base(t) = P_load(t) - P_onsite_generation(t)

For a non-exporting site, the required control action is:

P_control(t) = max(0, P_base(t) - P_cap)

The feasible cap is found by simulation. Sweep several cap values through the chronological data and record:

• number of control events;
• maximum control power;
• energy required per event;
• maximum event duration;
• consecutive events and recharge opportunity;
• curtailed production or comfort impact;
• battery throughput and expected operating range;
• residual billing demand under the exact tariff model.

A lower cap is not automatically better. It can increase event frequency, storage cycling, process disruption, and the risk of missing the cap after a forecast or communications error.

5. Use Predictive Interval Control

A controller that reacts only after measured demand exceeds the target may act too late. For an interval of duration (T), at elapsed time (t), let (E_{elapsed}) be imported energy accumulated in the current interval and (P_{now}) the estimated net import for the remaining interval. A simple projection is:

P_projected = [E_elapsed + P_now × (T - t)] / T

The remaining energy headroom under the cap is:

E_headroom = P_cap × T - E_elapsed

Use consistent units: if power is in kW and energy is in kWh, express (T) and (t) in hours.

If projected demand exceeds the cap, the controller dispatches available resources according to priority. In practice, a robust estimator should consider:

• revenue-meter interval synchronization;
• meter and communications latency;
• load ramp rate and forecast error;
• actuator response time;
• VFD minimum speed and process constraints;
• contactor switching duty and minimum on/off time;
• compressor, chiller, furnace, or pump restart restrictions;
• BESS state of charge (SOC), temperature, power limits, and reserve;
• hysteresis or deadband to prevent chattering;
• fallback behavior after sensor, network, or controller failure.

Do not use a switching contactor as a high-frequency modulating element. Use a VFD where a motor-driven process is suitable for continuous setpoint adjustment, and keep switching frequency within the equipment’s verified duty. The VIOX explainers on VFD operating principles ve contactor switching function provide the component-level background.

6. Build a Load-Priority Matrix

Every controllable load needs a documented operational contract. “Can be turned off” is not enough.

Alan	Required definition
Load ID and measurement point	Which feeder or process is being controlled and how its response is verified
Control mode	Start inhibit, schedule shift, staged off, setpoint reduction, or power limit
Available reduction	Expected kW as a function of operating state, not only nameplate rating
Yanıt süresi	Delay from command to measurable grid reduction
Minimum/maximum state	Safe minimum speed, temperature, pressure, flow, SOC, or output
Maximum interruption	Time before product, safety, comfort, or equipment limits are affected
Restart behavior	Inrush, warm-up, sequencing, rebound demand, and minimum off time
Öncelik	Never curtail, modulate first, defer, or interrupt only as a last resort
Fail-safe state	Result after loss of meter data, communications, controller, or actuator power

Safety systems, protection functions, emergency ventilation, life-safety loads, and loads whose interruption creates a hazardous process state are outside ordinary demand-shedding logic unless a competent design authority has explicitly integrated them.

7. Demand-Control System Architecture

A typical architecture has five functional layers:

1. Billing boundary: the utility revenue meter and active tariff determine the charge.
2. Operational measurement: a plant meter, correctly scaled CTs, submetering, and time synchronization provide faster data for control and diagnosis.
3. Demand estimator: an energy management system (EMS), programmable logic controller (PLC), or dedicated controller reconstructs interval demand and forecasts the interval end.
4. Dispatch logic: priority, process permissives, SOC limits, deadband, export constraint, and fail-safe rules determine the action.
5. Actuators: scheduling systems, VFD setpoints, managed electric vehicle charging, control relays/contactors, generator controls where permitted, and the BESS power conversion system (PCS).

Power protection remains independent of the optimization objective. The demand controller does not replace overcurrent protection, ground-fault protection, anti-islanding functions, emergency shutdown, or equipment-specific safety controls.

8. Size BESS Power and Energy Separately

Battery power and battery energy solve different constraints. The article kW, kWh, MW, MWh, SOC, and depth of discharge explains these terms in detail.

8.1 PCS power rating

For a time-series feasibility model, the required AC discharge power is:

P_BESS,AC(t) = max(0, P_base(t) - P_cap - P_flexible_load(t))

Here, (P_{flexible_load}(t)) is defined as a positive reduction delivered by scheduling, modulation, or approved shedding relative to the baseline.

The preliminary PCS discharge rating must be at least the maximum required value:

P_PCS,discharge ≥ max[P_BESS,AC(t)]

The final PCS specification must also account for overload capability and duration, temperature and altitude derating, reactive-power duty, import/export constraints, transformer and conductor capacity, fault contribution, harmonic performance, and the interconnection agreement. If the PCS must recharge within a defined window, charge power is a separate requirement.

8.2 Energy delivered during an event

For event (j), required AC energy is the area above the cap after flexible-load control:

E_AC,event,j = ∫ P_BESS,AC(t) dt

Convert the worst credible event to a preliminary nameplate energy:

E_nameplate ≥ E_AC,event / (η_dis × f_usable × f_EOL)

nerede:

• (η_{dis}) is the modeled discharge-path efficiency over the relevant operating range;
• (f_{usable}) is the permitted fraction of nameplate energy between operating SOC limits;
• (f_{EOL}) is the retained-capacity fraction required at the design end of life.

Include auxiliary loads and temperature effects in the time-series model when material. Do not stack an unexplained “safety factor” on top of hidden conservative assumptions; state each allowance once.

8.3 Chronology can govern the energy rating

A 15-minute demand interval does not mean a 0.25-hour battery is sufficient. A plant can remain above the cap for several consecutive intervals. It can also experience a second event before the battery recovers. Simulate:

• continuous multi-interval peaks;
• recharge power and permitted recharge times;
• recharge creating a new demand peak;
• minimum SOC reserve;
• forecast error;
• degradation over the analysis period;
• simultaneous reactive-power requirements if applicable.

NREL’s System Advisor Model battery documentation supports time-series dispatch, including peak-shaving controllers, and models temperature and degradation effects. That is the appropriate level of analysis for an investment decision; a single monthly maximum is not.

9. Worked Example: Derive PCS kW and Preliminary Battery kWh

Assume a simple 15-minute block-demand tariff, no ratchet, no export, and a candidate demand cap of 800 kW. After all approved flexible-load actions, five consecutive interval-average net imports are:

Ara	Baseline import (kW)	Required BESS discharge (kW)
1	720	0
2	860	60
3	980	180
4	910	110
5	760	0

Assume each interval remains at the listed average for the full 0.25 h. The minimum power requirement from this event is:

P_PCS,discharge ≥ max(60, 180, 110) = 180 kW

The AC energy delivered is:

E_AC,event = (60 + 180 + 110) kW × 0.25 h
           = 87.5 kWh

For an illustrative discharge efficiency of 0.94, usable SOC fraction of 0.80, and end-of-life retained-capacity fraction of 0.80:

E_nameplate ≥ 87.5 / (0.94 × 0.80 × 0.80)
            ≥ 145.4 kWh

The preliminary result is therefore at least 180 kW PCS discharge capability and 145.4 kWh nameplate energy for this simplified event and these assumptions. It is not a procurement specification. A chronological annual simulation may produce a longer event, inadequate recharge window, higher power ramp, lower efficiency, or operating reserve that governs the actual rating.

10. Model the Tariff Result Without Promising Savings

For a tariff with one simple monthly demand rate (R_m), an initial demand-charge model is:

Annual demand-charge reduction = Σ [R_m × (D_base,m - D_controlled,m)]

Do not use this expression unchanged when the tariff includes ratchets, multiple demand periods, coincident peaks, kVA demand, a contract floor, standby charges, or tiered demand rates. Rebuild the bill line by line.

The technical-economic model should include:

• controlled demand under each tariff register;
• energy shifted into different price periods;
• BESS round-trip losses and auxiliary consumption;
• battery degradation and replacement assumptions;
• maintenance and communications costs;
• production impact from load curtailment;
• demand-response revenue only where a signed program defines baseline and performance rules;
• residual demand risk after forecast or equipment failure.

Report sensitivity to tariff, peak shape, production schedule, battery degradation, and cap compliance. A universal savings percentage or payback period is not technically defensible.

11. Electrical, Interconnection, and Safety Boundaries

Demand reduction does not reduce the need for a full electrical design. At minimum, verify:

• service and transformer loading in both charge and discharge states;
• short-circuit current, protection coordination, grounding/earthing, isolation, and emergency shutdown;
• CT polarity, ratio, class, burden, meter accuracy, and loss-of-signal behavior;
• PCS voltage, frequency, harmonics, reactive-power mode, fault response, and anti-islanding requirements;
• export limitation and utility interconnection approval;
• enclosure location, ventilation or thermal management, access, clearances, and environmental conditions;
• battery management system limits, alarms, fire detection, commissioning, and emergency response;
• cybersecurity and access control for networked meters, EMS, PLC, EVSE, and PCS interfaces.

For U.S. projects, relevant scope references can include NFPA 855 for stationary energy storage installation, UL 9540 for energy storage systems and equipmentve IEEE 1547 for DER interconnection and interoperability. The adopted editions and applicability depend on the jurisdiction, system configuration, and AHJ. Other countries use different statutory and standards frameworks.

12. Commissioning and Measurement-and-Verification Plan

Commissioning should prove the complete control chain, not only that individual devices communicate.

Meter and model acceptance

• Verify CT orientation, scaling, time synchronization, and import/export sign.
• Compare operational-meter interval energy and demand with the utility data over representative periods.
• Confirm the controller resets and aggregates intervals according to the intended tariff model.
• Test missing data, frozen values, out-of-range readings, and clock failure.

Control acceptance

• Force a controlled test profile below the real demand limit where practical.
• Confirm projected demand, dispatch threshold, deadband, and release threshold.
• Measure actual kW reduction and response time for every flexible load.
• Verify minimum on/off time, restart sequence, rebound demand, and process permissives.
• Confirm that loss of communications or controller power produces the documented fail-safe state.

BESS acceptance

• Verify commanded versus measured AC power and energy.
• Test SOC limits, reserve, export restriction, charge recovery, alarms, and emergency stop.
• Confirm that recharge scheduling does not create a new chargeable peak.
• Demonstrate the longest modeled event or an approved equivalent test.

Billing verification

• Reconstruct each monthly bill using controlled interval data.
• Compare physical peak reduction with billed-demand reduction; investigate ratchet or tariff-floor effects.
• Track cap misses by cause: forecast, meter, communications, unavailable load, SOC, equipment trip, or operator override.
• Re-run the chronological model after material production, tariff, generation, or equipment changes.

The result of this process is not simply a smaller peak on a dashboard. It is a traceable chain from tariff definition to measured interval, control decision, physical load response, and verified billing outcome.

Kaynaklar

• DOE FEMP: Evaluating Your Utility Rate Options
• DOE Better Plants: Demand Response in Industrial Facilities
• ORNL: Understanding Your Utility Bills—Electricity
• NREL System Advisor Model: Battery Storage
• NFPA 855: Standard for the Installation of Stationary Energy Storage Systems
• UL 9540: Energy Storage System Testing and Certification
• IEEE 1547-2018: Interconnection and Interoperability of Distributed Energy Resources