Maximum Demand Monitoring: Catch Peaks Before the Bill

A 1,000 kW peak on a Medium Voltage General tariff costs roughly RM89,270 a month in capacity and network charges, and that whole number is set by one 30-minute interval. Not a week of bad operation, not an average: a single half hour. The awkward part is timing. The bill that tells you what your maximum demand was arrives four weeks after the interval that set it, long after the chiller staged up during a restart or the compressor and the oven landed on the same half hour. That is the case for maximum demand monitoring: measuring demand as it forms, on site, while an operator can still do something about it.

Why the monthly bill is useless as a control signal

The bill is an invoice, not a feedback loop. Under RP4, TNB bills Medium Voltage customers per kW of the month's single highest 30-minute demand interval: RM29.43 capacity plus RM59.84 network on General (RM89.27/kW/month), or RM30.19 plus RM66.87 on ToU (RM97.06/kW/month). By the time you read that figure, the interval is a month old. Nobody remembers what ran at 3:30pm on a Tuesday five weeks ago, and the logbook rarely says.

That makes post-hoc analysis a forensic exercise. You can reconstruct the peak from interval data if you have it, and that is worth doing, but reconstruction never prevents anything. Prevention needs the number in front of an operator during the interval itself. If you are still working out how your peak is derived in the first place, start with how to calculate maximum demand before wiring up alerts.

What maximum demand monitoring actually measures

It measures average kW over the current billing interval, not instantaneous kW. This distinction is where most in-house dashboards go wrong. A screen showing live instantaneous power is interesting, but it is not what TNB charges for. A 400 kW spike lasting 90 seconds barely moves a 30-minute average. A 200 kW step increase that stays up for the rest of the interval moves it a lot.

Practical maximum demand monitoring therefore tracks three things at once: energy accumulated so far in the interval, minutes elapsed, and the projected interval average if load continues unchanged. That projection is the alarm variable. In CobiNeural the Max Demand KPI does exactly this on the incoming supply, alongside the same view per feeder or per production line so you can see which board is driving the number rather than just that the number is rising.

How much warning does a 30-minute interval give you?

Enough to matter, if you are watching the projection rather than the total. Work an illustrative case: your threshold is 1,000 kW, so the interval budget is 500 kWh (1,000 kW for half an hour). Ten minutes in, the site has been drawing an average 1,200 kW, which is 200 kWh consumed. You have 300 kWh left for the remaining 20 minutes, which means the site must average 900 kW for the rest of the interval to land exactly on 1,000 kW.

That is a real, achievable instruction: shed roughly 300 kW for 20 minutes. And it was visible at minute 10, which is why alerting on projected demand typically buys 10 to 15 minutes of reaction time. Wait until minute 25 and the arithmetic turns hostile: with 5 minutes left and 480 kWh already spent, no reasonable shed brings the average back down. Early alerts are not a nicety, they are the whole mechanism.

Setting a threshold you are willing to pay for

Set the alert threshold below the peak you are prepared to buy, not at your historical maximum. Pick the kW ceiling you want to hold this month, then set the warning band a few percent under it so the alert lands with time on the clock. Sites commonly run two levels: an advisory at around 90 percent of the ceiling and an action alert at 95 percent or on projected breach.

Two refinements matter in Malaysian practice. First, on a ToU tariff, maximum demand recorded outside 2:00pm to 10:00pm on weekdays is not charged, so a demand alarm that screams at 6:00am is noise. Suppress or relax off-peak thresholds and keep operator attention for the window that costs money. Second, once the month's peak has already been set at, say, 1,180 kW, alerting at 1,000 kW for the rest of that month produces alarms with no financial consequence. The threshold should track the month's running peak, not sit frozen.

What the operator does in those 10 minutes

The alert is worthless without a pre-agreed response list. Write it before you need it, with named loads, expected kW, and who is authorised to touch them. On most commercial and industrial sites in Malaysia the list looks like this:

- Delay the next chiller stage. Let chilled water setpoint drift 0.5 to 1 degC for 15 minutes rather than bringing another compressor online mid-interval. Thermal mass is your buffer.
- Pause discretionary loads. Battery charging bays, non-critical air compressors topping up receivers, water transfer pumps with headroom in the tank, electric water heating.
- Stagger restarts after breaks. The 2:00pm return from lunch is a classic peak-setter: machines, HVAC recovery and lighting all step up together, right at the start of the ToU peak window. Sequence the restart across 10 to 15 minutes.
- Hold back a test run or a large batch start until the interval boundary passes.

Send the alert where the person actually is. CobiNeural Alerts push to WhatsApp and email, so the duty technician on the plant floor gets the same demand warning as the facility manager, with the current projection and the offending feeder in the message. The Dashboards then show whether the shed worked, in the same interval.

Do you need to replace your BMS to do this?

No. Most sites already have the measurement points: an incoming meter, a BMS trending chiller and AHU status, PLCs on the production side. What is usually missing is one place that computes interval demand correctly, holds the threshold logic, and notifies people. CobiNeural deploys standalone or as an overlay on existing BMS, PLC and SCADA, reading what those systems already produce rather than displacing them, which keeps the controls contractor out of the project.

Monitoring will not fix everything. A power factor below 0.85 (for supplies below 132 kV) still attracts a surcharge no alert can dodge, and a genuinely undersized off-peak window or a badly sequenced plant needs engineering, not notifications. For the structural side of the problem, see how to cut TNB maximum demand charges, and check current rates on the official TNB tariff schedule. But the recurring, avoidable peak (the one caused by two things starting at once, on a Tuesday, in a half hour nobody was watching) is exactly what live monitoring and a good alert threshold remove.

Want to know what this is worth on your site? Estimate the exposure yourself with our maximum demand calculator, or send us a recent TNB bill and a month of interval data and we will show you which half-hours set your peak and what a threshold would have caught. See how the CobiNeural platform handles it, then request a demo and we will run your numbers on the call.