If you manage a large industrial facility — food processing, automotive components, plastics, metals — demand charges are probably the most frustrating line on your electricity bill. The logic seems almost punitive: one 15-minute interval per month where your facility drew more power than any other 15-minute window sets your demand charge for the entire billing period. Consume that peak for 15 minutes in January and pay the penalty on every subsequent bill.
The demand charge structure exists for a real reason: utilities size their distribution infrastructure for peak load, and facilities that spike unpredictably force investment in grid capacity that sits idle most of the time. From the utility's perspective, charging for peak demand is rational cost recovery. From the facility manager's perspective, it is an incentive to avoid coincident peaks — and with the right tools, you can act on that incentive systematically.
Understanding the 15-Minute Demand Interval
Demand charges in most industrial tariffs are calculated based on the highest 15-minute average demand (kW) recorded during the billing month. Some tariffs use coincident peak (your demand during the utility's system peak hour, typically announced the next day) rather than non-coincident peak (your own highest interval regardless of grid conditions). Understanding which one your tariff uses matters for how you prioritize load shedding.
For a facility on a non-coincident peak tariff: your goal is to flatten your own load profile. Every kilowatt you cut from your single highest 15-minute interval saves you money, regardless of what the grid was doing at that moment.
For a facility on a coincident peak tariff: your goal is specifically to reduce load during the utility's system peak hours — typically hot summer afternoons in most U.S. territories. Reducing your own highest interval when it occurs at 3 AM in November is irrelevant to your demand charge under this structure.
The distinction matters enormously for optimization strategy. We have seen facilities invest in load-shifting measures that reduced their own peak by 15% but happened to shift load from non-coincident hours to coincident hours — their demand charge went up. Always analyze your tariff structure before designing a load management program.
The Industrial Load Profile: What Creates Demand Spikes
Industrial demand spikes typically come from a handful of identifiable sources:
Shift changeover and production start-up: When a production line restarts after a break or shift change, compressors, conveyor motors, and process heaters all come online within a short window. If start-up is not staggered, the simultaneous inrush can create a demand spike that takes only a few minutes to set the monthly peak.
HVAC coincidence with production peaks: In facilities with significant process heat loads, cooling systems often run at maximum capacity during peak production hours — exactly when process equipment is also drawing maximum power. This stacking of HVAC and production load is a common source of demand spikes that could be partially mitigated by pre-cooling before peak production begins.
Large motor starts: Variable frequency drives (VFDs) significantly reduce start-up current on larger motors, but many older facilities still have direct-on-line starters on equipment that pre-dates VFD cost reductions. A 100 HP pump started direct-on-line can draw 600–700% of full-load current during start-up — a brief but intense demand spike. Staggering these starts is a low-cost intervention with meaningful demand impact.
Compressed air system recovery: Compressed air systems that have been depressurized during a production break will run compressors at maximum output to re-pressurize the header. If this happens simultaneously with production start-up, the combination creates a spike. Pre-pressurizing the air system before the production line restarts eliminates this contribution.
Where Predictive Scheduling Applies
Predictive scheduling is most valuable for the category of loads that have temporal flexibility — where there is a window of time within which the operation must occur, but the specific start time is not fixed. The classic examples are compressed air pre-pressurization, HVAC pre-cooling, refrigeration defrost cycles, and some material handling operations.
For these loads, the approach is: given a forecast of the production schedule and the resulting expected demand profile, identify when the non-production loads will naturally want to run (or when they need to run based on process requirements), and shift them to windows where they add the least marginal demand.
Consider a food processing facility with a morning production shift starting at 6 AM. The compressed air system needs to reach operating pressure before production starts. HVAC needs to be at target temperature for worker comfort and food safety compliance. Both of these systems will draw significant power. The question is: do they draw that power simultaneously with each other and with the production line start-up, or are they staggered to flatten the demand profile?
A predictive scheduling system can calculate: given the production line start sequence (which equipment starts when), what is the expected demand profile from 5:30 to 7:00 AM? Then it identifies the pre-conditioning operations — compressed air pre-pressurization, HVAC ramp-up — and schedules them to fill the demand valleys before and between production starts, rather than stacking on top of them.
The result is a smoother demand curve over the 90-minute start-up window. The total energy consumption is nearly identical. The peak demand is significantly lower — often 10–20% reduction in 15-minute peak for facilities where production start-up previously drove the monthly peak.
What Predictive Scheduling Cannot Do for Industrial Demand
We are not saying predictive scheduling is a complete solution for industrial demand charges — there are categories of demand that cannot be shifted without disrupting production.
Process loads that run at a fixed rate determined by throughput requirements are not schedulable. If a smelter needs to maintain a specific temperature and power input to process material at the required rate, that load is process-constrained. You can optimize around it by shifting other loads away from its peak, but you cannot shift the process load itself without sacrificing throughput.
Demand charges driven by large motor starts (direct-on-line starters) are better addressed through equipment upgrades — VFDs, soft starters — than through scheduling. The scheduling window for avoiding a direct-on-line start inrush is zero seconds; there is no meaningful way to predict and pre-empt a spike that lasts for 200–400 milliseconds.
For facilities where process demand dominates the load profile and non-schedulable loads represent 80%+ of peak demand, the demand charge reduction from scheduling optimization will be modest — perhaps 5–8%. For facilities where HVAC, compressed air, and other infrastructure loads are a significant fraction of peak, the opportunity is larger.
The Data Requirements for Industrial Predictive Scheduling
Effective predictive scheduling for industrial demand reduction requires more telemetry than commercial building optimization. Specifically, you need:
Production schedule data: When do production lines start, stop, and change products? This is usually available from the manufacturing execution system (MES) or ERP system, but integration into an energy management platform requires an API or data export. In smaller facilities, it may mean a simple shift schedule that the facility manager enters manually or syncs from a scheduling spreadsheet.
Equipment state telemetry: Which large loads are running at any given time, and at what level? For multi-drive production lines, this may require integration with the facility's SCADA system. For simpler facilities, it may just be power consumption monitoring at the feeder level.
Process constraint information: Which loads cannot be deferred and by how much? This is typically entered as configuration rather than read from a live system — the facility engineer defines the constraints, and the scheduling system respects them.
The integration complexity is higher for industrial settings than for commercial buildings. BACnet and Modbus cover most HVAC and building automation needs. Industrial process systems often speak Profibus, EtherNet/IP, or OPC-UA. Getting a unified view of both building and process loads in one system is achievable but requires upfront integration work.
Measuring the Impact
Demand charge reduction is one of the most measurable ROI categories in energy management. Your monthly demand charge is directly stated on your utility bill, and the peak demand reading that set it is available from your meter data. Before-and-after comparisons over multiple billing periods give you a clean, auditable savings number.
The one complication: demand charges are affected by many factors, including seasonality, production volume, and equipment changes. To isolate the impact of scheduling optimization, compare demand charges in equivalent production-intensity periods before and after implementation, weather-normalized for HVAC-heavy facilities.
In our experience, facilities where HVAC and infrastructure pre-conditioning represent more than 20% of peak demand, and where production start-up stacking was previously unmanaged, tend to see 12–22% reduction in monthly peak demand readings over the first 6 months of predictive scheduling. Facilities with very tight process-constrained load profiles see less — 5–10% — but the baseline is often higher in absolute dollar terms, so the savings can still be significant.