Energy & load:
Forecast consumption, smooth peaks
Energy is not a fixed cost block but controllable. Time-series models forecast consumption per machine and shift – and smooth the expensive load peaks that determine your grid charge.
01 The problem – You pay for the highest quarter-hour
Energy cost arises from consumption AND load peak
Industrial energy cost has two drivers: total consumption and the load peak. A single quarter-hour in which several machines start up at once can determine the grid charge for a whole year.
Both are visible in the data: load curve per machine, shift plan, order load. A model forecasts the peak before it occurs – and suggests which start-up can be shifted by minutes without disrupting the plan.
Without loss of comfort and without touching production volumes.
02 The model – Forecast consumption per machine and shift
Time-series regression on load curve, production plan and outside temperature
Prognostizierte Lastspitzen-Fenster/Monat: 6 Spitzenlast: 4.180 kW -> 3.560 kW
Reacting only when the peak is reached is too late. The model forecasts the load curve per quarter-hour and detects hours in advance when several machines collide. Peak shaving becomes plannable instead of panicked – individual start-ups offset by minutes, the peak breaks.
03 Business impact – Cut consumption and peak load
The lever: reduced consumption plus smoothed peak-demand charges
No round number: every assumption comes from the sample plant and is stored centrally. With your real figures only the input changes, not the method.
| Item | Value |
|---|---|
| Energy cost / year | €3,200,000 |
| Consumption reduction (6%) | €192,000 |
| Peak-load cost / year | €480,000 |
| Peak smoothing (25%) | €120,000 |
| Result: savings / year | €312,000 |
Assumptions of a sample plant – in a real project your data replaces these values.
Consumption reduction (more efficient operation, avoided idling) and peak smoothing (peak shaving) draw from the same load-curve model. Peak smoothing often has the stronger effect because the demand charge enters the bill disproportionately.
04 Next steps
From the load curve to predictive control
We use existing meter and SCADA data per machine – the granularity usually already exists.
For one plant area the forecast model is built. You see the avoidable peaks of recent months back-calculated.
Predictive warning: "A load peak is due in 90 minutes – offsetting machine X start-up by 8 minutes is enough."