Blog
The plant paying more for electricity than it should, because nobody knows when the peak lands
Industrial electricity is not billed on units alone. It is billed on the fifteen minutes you used the most.
Published
Here is a question most plant managers cannot answer on the spot: last month, on what day and at what time did your peak demand occur, and which machines were running together at that moment?
It matters, because large industrial tariffs are not billed on total units alone. A demand charge is set by the highest consumption interval in the month.
Which means a single fifteen-minute window can determine a substantial part of a month's bill — and that window is usually caused by several machines happening to start at once, not by genuinely producing more.
Data held versus data used
Modern meters already record in short intervals. Plenty of plants have this data and use it only to check the invoice against, never to decide anything with.
Put it next to the production schedule and you can see what the peak is actually made of. The answer is often mundane: the air compressor starts at eight along with the office air conditioning and two ovens, when the ovens could start at half past without touching the production plan at all.
What forecasting adds
Knowing before you hit the ceiling. A model learning from historical consumption, combined with the day's schedule and the outside temperature, can estimate where load will be in half an hour. That is enough notice to delay starting something.
Planning production with energy in it. Once you know how cost varies by time of day and which jobs are energy-hungry, sequencing work becomes a decision with a cost dimension, not only a time dimension.
Spotting failures early. A motor drawing more current than usual at the same load is often on its way out. Consumption data is a maintenance signal you get for nothing, with no extra sensors.
What to watch for
Load forecasting is something models do well, but only as well as the inputs reflect reality.
A plant whose real schedule routinely differs from the one in the system will end up with a model that predicts the system accurately and the plant not at all. That is a process problem, and no amount of modelling fixes it.
Peak reduction also has a limit. If the plant already runs flat out around the clock, there is nothing to shift, and the answer is about equipment or self-generation instead — a different question entirely.
A straightforward way to start
Connect the meter data to something that keeps history, then do the simplest thing first: find out when the last ten peaks happened and what caused them.
That analysis takes a few weeks, and in many plants it pays for itself before any model is discussed.
What GIPSIC does here
We work on both sides of this — reading meters and sensors on site, and the systems that hold the history and put a usable screen in front of an engineer. We are a Google Cloud Partner and a Microsoft Solutions Partner for the parts that run in the cloud.
What we usually propose for a first project is a smaller scope than the client came in with, because proving the data is trustworthy is worth more than building something large on numbers nobody has checked.
If you would like to talk about energy at your plant, get in touch.
About the author
Film — Wisit. A businessman who still does his own BA work more often than he probably should, and writes a fair bit of code, front and back. Runs two or three small businesses. Follows technology and business obsessively, in Thailand and everywhere else. Off the clock: physics, astronomy, DIY, and anything to do with networks. Music always on, though he cannot sing. Plays instruments anyway, badly. Plays a lot of sport, racket sports above all. Not much of a traveller by himself, but happy to take Mint anywhere in the world. A man who fears — sorry, loves — his wife. One flaw: he barely touches video games.
Written with Claude Opus 5