Manaz Afzal
Manufacturing · Energy & Utilities

16% off the monthly demand charge, from a few hours of held load

Peak and power-factor management for an energy-intensive plant, measured on a full year of real 15-minute metering from a steel mill.

16%of the billed monthly peak removed
0.27%of energy moved to other intervals
7.5 hof held demand in a typical month
55% → 1%running intervals below 0.90 power factor

Executive summary

Holding the top 1% of 15-minute intervals cuts a steel plant’s billed monthly peak by 16% while moving 0.27% of its energy, measured on a full year of real metering.

16%of the billed monthly peak removed
7.5 hof held demand in a typical month
55% → 1%running intervals below 0.90 power factor

Situation

The monthly demand charge is set by a few coincident spikes, and the plant runs below a 0.90 power factor for much of its running time.

Approach

A demand monitor and cap per month, and reactive compensation sized from 35,040 real 15-minute intervals.

Result

16% off the billed peak for 7.5 hours of held load a month; intervals below 0.90 power factor from 55% to 1%.

The problem

An electric-arc or rolling plant pays a monthly demand charge on its highest interval of the month, often on kVA, so a poor power factor raises it further. Published work on steel plants shows that moving energy-heavy steps out of peak intervals and correcting power factor cut the charge noticeably without lowering output.

I built a monitor that watches interval metering against a monthly demand target and holds deferrable load for a few hours a month, and measured the saving on a real plant’s metering.

Team
Plant manager, electrical and maintenance engineers, production scheduling, energy procurement
Scale
960 MWh a year, 15-minute interval metering
Systems
SCADA and power-quality meters, the retailer’s interval data, MES or production scheduling
Measure
Billed monthly peak kW; intervals below 0.90 power factor; energy moved

What I built

A demand monitor that reads the interval meter every few minutes, projects the current 15-minute block and alerts the shift when it is heading above the month’s cap, with a short list of loads that can be deferred: furnace charging, compressor staging, a cooling or extraction cycle. Alongside it, a power-factor analysis sizes the reactive compensation needed for the plant’s real load pattern.

Interval meterkW and kvarevery 15 minutesProjectionWhere this blockwill landShift alertDefer a listed loadbefore the capMonthly reviewPeak, power factorand moved energy
The cap is set per month from the plant’s own data and agreed with production, so output is never held back.

Results

I worked with a full year of real 15-minute metering from a steel plant: 35,040 intervals, 960 MWh, with active and reactive energy, power factor and the plant’s load-period classification. The monthly peak sat on average 50% above the month’s 95th percentile interval, so it was set by short coincident spikes rather than by sustained production.

Januarypeak 613 kW, held to 552 kW10%Februarypeak 582 kW, held to 512 kW12%Marchpeak 605 kW, held to 483 kW20%Aprilpeak 556 kW, held to 456 kW18%Maypeak 560 kW, held to 484 kW14%Junepeak 535 kW, held to 448 kW16%Julypeak 487 kW, held to 440 kW10%Augustpeak 535 kW, held to 447 kW16%Septemberpeak 510 kW, held to 459 kW10%Octoberpeak 558 kW, held to 475 kW15%Novemberpeak 629 kW, held to 495 kW21%Decemberpeak 597 kW, held to 440 kW26%
Holding the top 1% of intervals to the month’s 99th percentile removes 16% of the billed peak on average, while moving 0.27% of the month’s energy into other intervals. Data: Steel Industry Energy Consumption, DAEWOO Steel, UCI Machine Learning Repository
Cap on 15-minute demandBilled peak removedEnergy moved to other intervalsTime held each month
Month’s 99th percentile16%0.27%7.5 hours
Month’s 98th percentile21%0.70%15.0 hours

That cap is reached in a median of 7.5 hours a month, a small enough window for the shift to defer a listed load each time. At the 98th percentile the cut grows to 21% for 15 hours of held demand a month.

0 kW100 kW200 kW300 kW00:0004:0008:0012:0016:0020:00WeekdayWeekendHour of day
Weekend demand runs well below weekday demand, which leaves room to move deferrable energy out of weekday peaks.

Power factor

55% of the intervals in which the plant drew power ran below a 0.90 lagging power factor, carrying 41% of its energy. The problem sits mostly in light-load periods, when motors and transformers idle. An automatic capacitor bank sized to the 95th percentile of the reactive need, about 81 kvar, brings that share down to 1.1% and lifts the year’s average power factor from 0.90 to 0.99.

As meteredWith capacitor bankLight-load periods2%Maximum-load periods1%Medium-load periods0%
The bank switches in steps as the reactive need changes, so it does not over-correct when the plant idles.
A day-ahead forecast of daily energy, fitted on January to September and tested on October to December, came within 30% of actual consumption against 35% for the same day last week. The gain is modest because output follows the order book, so I use it for procurement planning rather than for the peak alert.

Data and method

Data
Steel Industry Energy Consumption, UCI Machine Learning Repository: 15-minute metering from DAEWOO Steel in Gwangyang, South Korea, 2018-01-01 to 2018-12-31, with energy, lagging and leading reactive energy, power factors, CO2 and load-period type.
Peak cap
Each month’s 99th and 98th percentile 15-minute demand; energy above the cap counted as moved to later intervals
Power factor
Reactive need to reach 0.95 in each interval; an automatic bank sized at the 95th percentile of that need
Forecast
Gradient-boosted regression on day of week, month and lagged daily energy; fitted January to September, tested October to December
Measures
Billed peak removed, energy moved, hours held, intervals below 0.90

Limitations

How I would take it live

Financial impact

About A$14,878 a year off the demand charge

The gain is the lower billed peak each month. The demand tariff is an assumption; the peak cut is measured.

InputValueSource
Average monthly billed peak564 kWMeasured 2018 metering
Peak removed16%Measured 99th percentile cap
Demand charge per kW a monthA$14 network tariff varies
Energy moved0.27%Measured

Calculation

  1. kW removed from the monthly peak89 kW
  2. Demand charge avoided a yearA$14,878

Annual valueA$14,878

Power-factor charges avoided and freed supply capacity are further gains and are not counted here.

Implementation and rollout

Six weeks from interval data to managed peaks

  1. Week 1

    Data and tariff

    Twelve months of interval data and the current network tariff.

    Energy manager, Manaz

  2. Week 2

    Deferrable loads

    List of loads that can wait agreed with production and maintenance.

    Production, maintenance

  3. Weeks 3 to 4

    Advisory month

    Projection and alerts run on the historian; shifts log what they deferred.

    Shift supervisors

  4. Weeks 5 to 6

    Compensation and review

    Power-quality survey and capacitor specification; monthly peak review begins.

    Electrical engineer, Manaz

Tools

SCADA historian or power-quality meters, Retailer interval data, Python, scikit-learn

Team

  • Plant manager owns the cap and the load list
  • Electrical engineer reactive compensation
  • Shift supervisors act on alerts
  • Manaz monitor, analysis and monthly review

The timeline is a plan for a typical engagement; it depends on data access and how many systems are involved.

Risks and mitigations

What could go wrong in production, and the control for each

RiskMitigation
Output affectedOnly loads production signs off are deferred.
Over-correction at light loadAutomatic bank switched in steps.
HarmonicsPower-quality survey before specification.
Tariff changesCap and value recalculated on each tariff review.

Key takeaways

Key takeaways

  • A few intervals set the monthly demand charge.
  • Holding them cuts the billed peak by 16%.
  • Power factor is a light-load problem with a known fix.
  • Daily energy forecasts help procurement more than peak control.