How to Reduce Peak Demand Charges

Every demand reduction project is an attempt to do one thing: lower the highest interval average of the billing period without lowering the output of the business.

That framing rules a lot of things in and a lot of things out. It rules in anything that moves load, staggers it, stores it or supplies it locally. It rules out most of what is normally called energy efficiency, unless the efficiency measure happens to act during the interval that sets the charge.

The area under the two curves is similar. The height is not, and a demand charge prices the height.

Diagnose before you buy

The measures below are ordered by cost. Which one is right for a site is not a matter of taste — it is determined by the shape of the load, and the shape can be read from interval data in an afternoon.

Three numbers decide it.

The gap between your highest interval and your tenth-highest. This is the most important diagnostic on the site. A wide gap means very few intervals are setting the charge, so a cheap, occasional, targeted intervention captures most of the value. A narrow gap means the peak is essentially the whole load, and no amount of clever scheduling will help — you are looking at storage, generation or a different tariff.

The timestamps of the monthly maxima. If they cluster on the same weekday and hour, the peak is systematic and therefore controllable. If they scatter unpredictably, it is incidental, and any control strategy has to be automatic rather than procedural.

Your load factor. Energy divided by peak times hours in the period. High load factor sites have little shape to fix; low load factor sites usually have a great deal.

Getting the data to compute all three is covered in how to get your interval data, and it is genuinely worth doing first. The most expensive mistake in this field is buying a measure that addresses a peak the site does not have.

One more thing is worth settling before any measure is chosen, and it is not a property of the load at all. Establish whether the charge you are trying to reduce is measured against your own peak or against the system's.

A non-coincident charge bills your own highest interval, whenever it happened. It is visible in your data, attributable to specific equipment, and controllable. A coincident charge bills whatever you were drawing during somebody else's maximum, which you have to forecast rather than measure. Sites that spend capital flattening a peak the tariff was never billing are not rare, and the mistake is entirely avoidable by reading which determinant is which before starting.

Many bills carry both, at separate rates, and a measure that helps one may do nothing for the other.

Rung 1 — Sequencing and staggered startup

Cost: usually nothing. Typical finding: large.

The classic industrial peak is not a production peak at all. It is the moment after a break, a shift change or a weekend, when every compressor, chiller, air handler and conveyor is switched on within the same few minutes. The plant is not producing anything yet and the meter records the highest average of the month.

Three loads of 200 kW that start one after another never produce an interval average above 200 kW. The same three starting together produce 600 kW. Nothing about the equipment changed; the demand determinant tripled.

Fixing it means adding delays to a control program, or writing a startup order into a procedure and enforcing it. It is the first thing to check and it is covered in staggered startup.

Rung 2 — Scheduling and setpoint strategy

Cost: low. Requires: knowing when the window is.

Once simultaneity is dealt with, the next question is whether flexible load can be moved out of the expensive period entirely. Battery charging, non-critical pumping, tank heating, compressed air storage, test rigs, laundry, some batch processes — these frequently have more timing latitude than anyone has ever been asked to exploit.

On the cooling side, pre-cooling is the standard move: run the plant harder before the peak window and let the building coast through it on stored thermal mass. It is covered in pre-cooling and HVAC scheduling against the peak.

The caution here is the one that catches most projects. If your tariff bills both an on-peak demand and a facility demand, moving load out of the window reduces the first and not necessarily the second, because the facility measure simply finds the new peak wherever it now sits. Model both determinants against the modified load series before promising a number: facility, on-peak and billing demand.

Rung 3 — Demand limiting controls

Cost: moderate. Requires: loads that can actually be shed.

A demand limiting controller watches the accumulating interval average and sheds pre-agreed loads when it predicts the target will be exceeded. It automates what a very attentive operator would do, at a speed no operator can sustain.

It works well where there is genuinely deferrable load — cooling with thermal inertia, battery charging, non-critical process heating — and badly where there is not. A controller with nothing to shed simply raises alarms. The failure modes are worth reading before specifying one: demand limiting controls and how they fail.

Rung 4 — Thermal storage

Cost: significant capital. Suits: large, predictable cooling peaks.

Make cooling at night when demand is low, store it as chilled water or ice, and draw on it during the day. The cooling load is served; the electrical load that produces it has moved to a period where it costs less.

This is a well-established approach with a specific profile: it suits sites where cooling is a large fraction of the peak and the peak is predictable enough to plan against. It is treated in thermal energy storage for demand reduction.

Rung 5 — Batteries

Cost: high. Suits: peaks that are short, sharp and frequent.

A battery discharges into the peak and recharges outside it. It is the most flexible measure available — it does not care what the load is, only how large and how long — and it is the most capital-intensive.

The sizing question is where most battery projects are won or lost, because power rating and energy capacity are two separate specifications and a battery correct on one and wrong on the other fails. That arithmetic is worked in full in sizing a battery for peak shaving.

Rung 6 — Generation and curtailment agreements

Cost: varies. Suits: sites that can genuinely reduce load on request.

On-site generation displaces grid draw during the peak. Demand response programs pay you for the ability to reduce load when the system needs it, which for a site that already has a curtailment plan can be a second revenue stream on capability it has already built. What you are actually agreeing to is worth reading carefully: demand response programs.

What the ladder is worth

Every rung produces the same unit of value — an avoided kilowatt — and that unit has to be priced correctly or the comparison between rungs is meaningless.

The comparison is deliberately unfair, because that is the point: the cheap rungs are cheap. A site that installs storage before it has sequenced its startups has bought an expensive solution to a free problem — and worse, has sized the storage against a peak that need not have existed, which makes the equipment larger and the payback longer than it had to be.

There is a second reason to work upward rather than jumping. Each rung produces evidence. A sequencing change that visibly lowers next month's demand is a demonstration, in the site's own data, that the mechanism works as described — which is exactly the argument needed when a larger request goes to the capital committee. Projects that start at the top have to make that argument from a vendor's spreadsheet instead.

That said, the cheap rungs run out. Sequencing can only be done once. Scheduling is limited by how much genuine timing latitude the process has. When those are exhausted and a peak remains, the higher rungs need proper appraisal rather than a payback number — including tax treatment, incentives and the ratchet interaction — and that is what building the business case for demand reduction is for.

The multiplier most projects leave out

If your tariff has a ratchet, every calculation above understates the value, frequently by a factor of several.

Without a ratchet, avoiding a peak saves one month of demand charge. With one, it also avoids the floor that peak would have set under every subsequent month in the look-back window. The arithmetic is in ratchet clauses and the pricing in what a kilowatt of avoided peak is actually worth.

A ratchet also changes the objective. It stops being reduce the average peak and becomes never have a bad month, which favors reliability over maximum reduction. A measure that cuts 10 percent but permits one excursion a year is worth less than one that cuts 5 percent and never permits any.

What does not work

It is worth stating the negative case explicitly, because a substantial amount of money is spent on measures that cut energy and leave the demand charge untouched.

Anything that reduces consumption at hours other than the peak interval has no effect on the demand charge at all. Lighting that was already off, equipment that was already idle, night-time improvements at a site that peaks at noon: all of these produce real energy savings and zero demand savings. Six specific examples are in six efficiency measures that do not cut your demand charge, and the underlying distinction in load shifting is not energy efficiency.

The order to do it in

1. Get twelve months of interval data at the billing interval.
2. Compute the three diagnostic numbers. Decide whether the peak is systematic or incidental, and whether it is narrow or broad.
3. Read the tariff and establish which determinants are billed, whether a ratchet applies, and whether any charge is coincident.
4. Price one avoided kilowatt properly, riders included.
5. Work up the ladder. Do not skip a rung because the next one is more interesting.
6. Measure afterward, against the baseline you kept, in a way finance will accept: measurement and verification.

Step six is the one that determines whether there is a second project. A saving that cannot be demonstrated is, for capital allocation purposes, a saving that did not happen.