Objective
The electricity bills for industrial and commercial users include a demand component (maximum demand). By staggering the startup of air conditioners, air compressors, and other equipment, the monthly demand can be reduced. IoT gateways collect branch power, and in combination with the strategy engine and grid price signals, controls or prompts are issued via cellular or Ethernet. Implementation must ensure production process safety and avoid blind load shedding.
Communication
The real-time requirement is moderate, with minute-level scheduling sufficient; in case of network disconnection, it should fall back locally to safe settings. When interfacing with BMS or PLC, attention should be paid to electrical isolation and permissions.
Benefit assessment
It should be simulated based on historical load curves to quantify cost savings and equipment lifespan impacts, and provide customers with auditable reports.
Edge strategies and cellular link reliability
Load management instructions relate to production safety, so the gateway should locally store interpretable rules: when communication is interrupted beyond a threshold, it should automatically revert to conservative mode, for example, only issuing alarms without load shedding or unloading step by step according to preset priorities. 4G/5G links should distinguish between control plane and data plane timeouts to avoid false triggers. For critical circuits, dual-channel redundancy (cellular plus wired) or local PLC interlocks are recommended. Strategy issuance should include version numbers and signatures to prevent tampering; execution results and on-site instrument readings must be feedbacked for reconciliation, and any discrepancies should immediately trigger shutdown protection.
Data quality and project delivery
If there are errors in CT ratio, reversed phase sequence, or excessive harmonics in power collection, it may cause strategy misjudgment. Baseline calibration and a continuous seven-day comparison should be performed before commissioning. When delivering to customers, it is necessary to explain the controllable load list, minimum control granularity, and uncontrollable process boundaries, and obtain written authorization. Cost-sharing contracts often require third-party metering or platform log audits; the IoT platform should retain the original sampling and strategy decision chain to support dispute arbitration.
Engineering Practice Supplement
Cross-team collaboration for 'load management': Hardware provides impedance and power budget, RF provides spurious and sensitivity margin, embedded systems specify thread models and storage lifespan, cloud defines SLA and rate limiting. For scenarios like NB-IoT, the business model must match wake-up cycles, packet length, and downlink capacity; for vehicular and mobile scenarios, session continuity during handover and roaming must be considered. At project closure, conduct a 'red team' style review: check if weak network, power loss, certificate expiration, clock jumps, and full storage all have defined behaviors.
Demand windows and circuits that may be shed
Load management is aimed at maximum demand on the electricity bill. The meter takes the peak of average power inside a sliding window, and that peak is part of the charge. The control target is not “use one kilowatt-hour less.” It is to keep several large loads from sitting at rated power inside the same window. An IoT gateway reads branch power and incoming power and, by a priority written in advance, delays a start or briefly sheds a circuit that is allowed to stop, such as some air conditioning, a compressor, or a charger, when the demand threshold is about to be touched. A circuit the process cannot stop must be marked uncontrollable. The policy engine has no right to open it.
A project can proceed only with branch metering, actuators that can actually be operated, a person who has confirmed which loads may be touched automatically, and a local fallback that stops shedding or only alarms when the link is down. Turning on automatic shedding without a historical load curve cannot tell whether the saving came from shifting peaks or from cutting production. Actuators and the power circuit need galvanic isolation and manual override. A communication fault must not leave “close” or “open” as an ambiguous default.
Shedding a process load after the link is lost
A wrong current-transformer ratio, or a reversed phase order, makes calculated power negative or too small. The policy then stays idle while the real load is already high, or it sheds during a light load. Watching only the main meter, and not the process state, stops the compressor in the middle of a critical step. A command with no version and no signature is executed late when communication returns. Cellular jitter is treated as a reason to shed immediately, so a brief dropout opens a load. A savings report that uses the platform’s estimated energy does not match the utility’s settlement meter, and a revenue-share dispute has no raw samples to trace.
After the local controller has missed heartbeats for longer than the agreed time, it should enter the conservative state written down in advance, not hold the last aggressive command. Time-of-use price and demand are two rules. Tariff periods are covered in time-of-use metering. Do not replace the demand window with peak-and-valley price logic. If the site also follows a grid dispatch signal, look separately at the event timing in demand response, so two systems do not send opposite commands to the same circuit.
Before commissioning, check transformer ratios and phase order, and compare a reading against a known load. Hard-code uncontrollable circuits as forbidden to shed. Write the action after a communication loss as alarm or hold, not as “keep executing the last aggressive command.” Version each command, and do not replay an expired shed when the link returns. Keep raw samples and the reason for each decision for the settlement period, so a report that disagrees with the utility meter can be checked. Configure the demand window and the peak-valley price separately, so off-peak price logic is not used to cut a process that is underway.
Rehearse the local state after a loss of contact: when communication stops, does shedding actually stop, and does it catch up after the link returns? Store the rehearsal record with the policy version. If a branch power reading is negative, stop the policy and check wiring and ratio. Do not keep automatic control on a bad measurement. Do not open a circuit that is outside the controllable list, even when communication is healthy.