For an energy-intensive SMB, half the bill comes from peak-hour consumption. The classical energy OR problem: demand response, peak shaving, time-of-use scheduling.
In plain words
Sound familiar?
- 30-50% of our monthly electricity bill is connected-power โ 1-minute peak in demand adds 50K TRY/month.
- We have a cold store or ice room; compressors run 24h while the night tariff is half of day.
- Production has melting or heating loads; should we run only day or could we run nights too?
- We invested in or plan a rooftop solar; we don't know how to match the generation curve to consumption.
- We have a battery or plan to buy one; charge/discharge decisions are intuitive.
- We picked a free-market electricity tariff or need to evaluate one; we can't pick a tariff without producing a profile.
- When the bill arrives we say 'this month the peak look was high' but can't trace where it came from.
Why it matters
How it's solved
Technical depth
How it's solved
Technical depthOne-liner: To cut the bill, measure two separate things โ for the energy component (kWh ร tariff), shift loads to off-peak hours; for the connected-power (kW), slow the peak moment (don’t let big loads start together). Mixing the two is the common mistake.
This problem sits at the intersection of operations research (a discipline that uses math and computing to solve business-decision problems) and energy systems. Two core sub-problems: Demand Response Optimization (which load to run when under a time-of-use tariff), Peak Shaving (minimize the daily/monthly maximum demand). Solution in three stages:
1. Modeling. Load inventory: per device/process โ power (kW), hourly energy (kWh), flexibility type (shiftable, curtailable, fixed), constraints (line sequence, temperature, shift). Tariff: time-of-use (T1/T2/T3 โ day, peak, night), connected-power (kW), energy (kWh), capacity charge. With renewables: hourly generation forecast (solar irradiance). With a battery: capacity (kWh), charge/discharge rate (kW), efficiency, cycle life.
2. Solver-driven decision. Single-day plan: MIP โ hourly decision variables (each load on/off, battery charge/discharge level). 24-hour MIP solves in minutes on modern solvers. For multi-day or stochastic (uncertain renewable forecast), Stochastic MIP or Model Predictive Control (MPC โ a control method that updates its plan each hour based on new data). Industry-common software: GAMS / AMPL with custom energy models; Energy Management System (EMS) solutions.
3. Field integration. Output: hourly load schedule. PLC / SCADA direct integration (semi-automatic control) or tablet/printer for operator. Smart meter data closes the loop: planned vs. actual peak, actual tariff cost. Monthly energy report and KPI dashboard. With renewables/batteries, real-time control (MPC) is usually needed."
Alternatives
Manual + smart meter + operator experience
FreeSmart meters provided free by regulator
Who it fits: Small facility (<100 kW), simple tariff
- + Zero software cost
- + Operator decides which load to run when
- + Intuitive savings 5-10%
- โ Not optimal
- โ Hard to integrate battery or solar
- โ If peak hours don't shift, limited benefit
- โ Doesn't scale to many loads
Local Energy Management System (EMS) software
Enterprise30,000โ120,000 TRY license + 6,000โ20,000 TRY/year maintenance (TR market observation)
Who it fits: Medium energy-intensive SMB (100 kW-2 MW)
- + Local-language support, regulator-compliant reports
- + PLC/SCADA integration
- + Basic tariff-based optimization
- โ Usually no MIP / stochastic optimization
- โ Weak on battery + solar joint optimization
- โ Basic forecasting module
International industrial EMS + DR software
Enterprise40,000โ250,000 EUR license + 10,000โ40,000 EUR/year
Who it fits: Large energy-intensive facility (2+ MW) or campus
- + MIP / MPC-based optimization
- + Renewable + battery + load joint
- + Advanced demand-forecasting module
- โ Expensive
- โ 6-12 month deployment
- โ Local-regulator tariff structure customization needed
Open-source solvers + custom energy model
Open SourceLicense free; in-house build 12-20 weeks or 300K-900K TRY consulting
Who it fits: SMB with technology team and very specific constraints
- + No license
- + Mature open-source MIP solvers sufficient
- + MPC and stochastic optimization open-source
- โ In-house OR/energy expert required
- โ PLC/SCADA integration separate project
- โ Ongoing maintenance
Recommendation
Ask in the meeting
- Do you use MIP or MPC for demand-response optimization, or is it rule-based?
- How is the regulator's tariff structure (time-of-use + connected power + energy) represented in your model?
- Can you jointly optimize battery + solar + flexible load, or are they separate modules?
- Which method does your demand-forecasting module use (ARIMA, ML, seasonal)?
- Which protocols for PLC / SCADA integration? OPC-UA, Modbus, MQTT?
- Is the peak-shaving objective daily, monthly, or annual?
- Can we run a 4-8 week pilot with real consumption data and our current tariff during the trial? How are savings measured?
- If we stop working with you, how do we get hourly consumption data, plan history and meter data back?
Technical details
Editor’s note
Colloquially this is called “electricity savings”, “tariff savings”, “peak demand reduction”. Academically it goes by Demand Response Optimization, Peak Shaving, Load Shifting Problem. Without knowing these distinctions, you can’t tell the difference between an EMS that ‘monitors energy’ (shows smart-meter data) and an EMS that ‘optimizes energy’ (decides load schedules). The first only measures, the second saves.
The most-skipped point in the industry: modeling the tariff structure. The regulator’s tariff has four components: energy charge (kWh, time-based), capacity charge (connected power, kW), transmission/distribution, and taxes/fees. Each is optimized differently. If the software only minimizes kWh cost (ignoring connected-power charge), the intuitive savings number won’t translate to the actual bill. The tariff module is the critical test.
Step-by-step path โ for the SMB
Stage 1 โ Measure first, plan after. At least 8 weeks of smart-meter data (15-min or hourly granularity): total consumption, peak-hour consumption, load profile (hourly), tariff breakdown, current cost detail. Build a load inventory โ which device/process consumes how much (sub-metering or estimate).
Stage 2 โ Extract the knowledge asset. Which loads are flexible (shiftable, curtailable, conditional), which are critical. Production flow constraints (shift, sequence, temperature). Tariff selection (are we a free-market consumer, which tariff?). If renewables or battery: capacities.
Stage 3 โ Pilot. 8-12 weeks. Start with 1-2 flexible loads (e.g., cold-store compressor, laundry). Success criteria defined in advance: monthly bill -8% minimum, no production disruption.
Stage 4 โ Roll-out. 3-6 months for all flexible loads. PLC/SCADA integration with semi/fully automatic control. Monthly energy report and KPI tracking. With renewables/battery, add real-time control (MPC).
Risks โ what can go wrong
- Production disruption. Shifting loads must not break production or service quality. Flexibility constraints (temperature range, shift, sequence) must be entered correctly; monitor tightly in the pilot.
- Tariff changes. Regulator tariffs update once or twice a year. If the software doesn’t adapt, optimum is broken. Contract must define tariff-update process.
- Forecast error. Renewable-generation forecast error is 15-25%; battery and flexible-load decisions are sensitive to this. MPC or stochastic approach needed.
- Investment double counting. Battery system amortization must be netted against savings; reporting only savings is misleading.
Solution method โ a technical look
| Approach | Typical scale | Solve time | Guaranteed optimum? |
|---|---|---|---|
| Rule-based (off during peak, on off-peak) | Small site | instant | No, 50-70% optimum |
| MIP (deterministic 24h) | Medium-large site, 10-50 loads | minutes | Yes (within limit) |
| Stochastic MIP (renewable uncertainty) | Renewable + battery + load | hours | No, 95-98% |
| Model Predictive Control (MPC) | Real-time (rolling 24h) | minutes/hour | Practically good |
| Reinforcement Learning | Very complex system | Training weeks | No, experimental |
Objective function choice:
- Objective 1 โ Total bill minimum: Standard. Includes both energy and connected-power.
- Objective 2 โ Peak power (kW) minimum: When connected-power dominates.
- Objective 3 โ Self-consumption maximum: For owners of renewables + battery.
- Objective 4 โ Carbon footprint minimum: For environmentally-focused sites."
Multi-objective: weighted sum or hierarchical. If ancillary-market participation exists (TR EPDK), that revenue is also added."
Academic references
See the sources field in the frontmatter.
Sources
- Albadi, M. H. and El-Saadany, E. F. (2008). A summary of demand response in electricity markets. Electric Power Systems Research, 78(11), 1989โ1996. Survey reference for DR literature.
- Palensky, P. and Dietrich, D. (2011). Demand side management: Demand response, intelligent energy systems, and smart loads. IEEE Transactions on Industrial Informatics, 7(3), 381โ388. Modern reference for industrial DR.
- Boyd, S. and Vandenberghe, L. (2004). Convex Optimization. Cambridge University Press. MIP/MPC theoretical foundation.
- Turkish Energy Market Regulatory Authority (EPDK) โ Electricity Tariff Regulation. Reference for the TR energy market.
- Turkish CoHE (YรK) Thesis Center โ keywords: ’talep tarafฤฑ yรถnetimi’ or ’tepe yรผk’ โ 40+ Turkish theses. tez.yok.gov.tr
Glossary
- Demand Response
- An energy-management approach that optimizes when loads run on the consumer side, based on the tariff, system conditions, or incentives.
- Peak Shaving
- Deliberately reducing the daily or monthly maximum power draw (kW) to lower the connection-power fee.