Pull the interval data for a manufacturing site and rank the intervals. Then look at the timestamps of the top ten.
At a surprising number of plants, most of them are within twenty minutes of a shift start, a break ending or a Monday morning. Not during peak production. Before production, when the plant is doing nothing but waking up.
Why the startup peak is so large
Steady-state operation is naturally staggered. Machines cycle, compressors unload, chillers stage down, batches finish and start at different times. Averaged over fifteen minutes, running production is smoother than it looks.
A cold start is the opposite. Everything is off, then everything is switched on, and for a period the site draws:
- Full load on equipment that will later cycle. Compressors run continuously to build pressure in an empty receiver. Chillers run flat out to pull down a warm loop. Neither will do that again for the rest of the day.
- Heating elements at full duty. Ovens, tanks and process heaters draw their full rating during warm-up and then modulate to a fraction of it.
- Simultaneous operation of loads that never normally coincide. In production, load A runs while load B waits. At startup, both run.
The result is an interval average that exceeds anything the site produces during actual work — recorded, billed, and under a ratchet, remembered.
The arithmetic
Simultaneity is the whole mechanism, and it is worth seeing in numbers.
Three loads, two sequences
Same equipment, same energy, different demand determinant.
- Compressor, startup load210 kW
- Chiller, pulldown load260 kW
- Process heater, warm-up load180 kW
- Base load, everything else340 kW
- (All three started together, 15-minute average)990 kW
- (Started 8 minutes apart, worst 15-minute average)690 kW
- (Demand avoided)300 kW
- All-in value of an avoided kW-month$19.05
Annual value of the sequence change$68,580
The equipment is identical, the energy consumed is identical, and the startup finishes at almost the same time. Only the order changed. Figures illustrative.
Nothing was bought. Nothing was turned off. The plant reaches production readiness within a few minutes of when it would have anyway.
How to build the sequence
- Inventory every load above roughly 50 kW that participates in a cold start, with its starting draw and the time it takes to settle to normal operation.
- Establish which loads must be running before others can start — pressure, temperature and safety interlocks come first and are not negotiable.
- Establish which loads have real latitude. Space conditioning and non-critical heating usually do; process interlocks usually do not.
- Order the sequence so that no two large loads are in their high-draw phase at the same time, and so the fifteen-minute average never contains more than the planned number.
- Start the whole sequence earlier rather than finishing it later, so readiness time is unchanged.
- Implement it in the control system where possible, so it happens whether or not anyone remembers.
- Verify it against the next month's interval data — the startup peak should be visibly lower and the plateau slightly wider.
The distinction in the third and fourth steps carries most of the work. Sequencing is constrained by process interlocks, not by preference, and the sequence has to be designed with someone who knows the process. The point is to find the latitude that does exist, which is usually more than the first conversation suggests.
Where it goes wrong
It is written down but not enforced. A procedure that depends on an operator remembering the order at six on a Monday morning will be followed most of the time, and most of the time is not enough — one missed sequence sets the month's peak, and under a ratchet, the floor for the year. Put it in the control system.
Only the obvious loads are staggered. Sequencing the chillers and leaving eight air handlers on a shared optimum-start schedule leaves a large simultaneous block untouched. Air handlers, pumps and battery chargers count.
The restart after an outage is not covered. A site that sequences its Monday start and reboots everything simultaneously after a power interruption has a sequence that works on the day it is not needed. Automatic restart logic needs the same staggering.
Nobody checks whether it worked. The verification step takes ten minutes with the next data extract, and skipping it is how a sequence that was quietly disabled during a controls upgrade goes unnoticed for a year: how to get your interval data.
The events nobody plans for
Startup peaks also come from occasional operations that are outside the daily routine, and these are the ones that set ratchet floors because nobody was watching:
- Commissioning a new line while the existing plant runs normally.
- Restarting after a shutdown or a holiday, when the building is cold or hot and everything runs at full duty for longer.
- Annual maintenance tests that deliberately run equipment simultaneously to prove capacity.
- A supplier demonstration, or a factory acceptance test.
Each of these deserves the same treatment as a routine start: a sequence, agreed in advance. Where a ratchet applies, a single one of these can be the most expensive quarter of an hour of the year — ratchet clauses sets out how far that reaches.
Where it sits
Sequencing is the first rung of the ladder in how to reduce peak demand charges, and it is deliberately first because it is nearly free and frequently large. A site that installs storage before sequencing its startups has bought an expensive answer to a free question — and worse, has sized the storage against a peak that need not have existed, which makes the storage larger and the payback longer: sizing a battery for peak shaving.
Where the load being staggered is a chiller or an air handler, sequencing overlaps with scheduling proper, and the two are usually designed together: pre-cooling and HVAC scheduling against the peak.
