A plant on CFE's Gran Demanda en Media Tensión Horaria tariff pays for electricity three ways at once. It pays for energy at a different rate in base, intermedia and punta hours. It pays a capacity charge set by its maximum demand during the punta period. And it pays fixed, distribution and system charges that do not move with consumption at all. Software that shifts load out of punta changes the first line. Software that shaves the coincident peak changes the second. Neither one touches the third.

That is the basis on which an energy management platform has to be judged. It is also why a percentage saving quoted without a line item is not a number a finance director can use. A cut in kilowatt hours does not cut a GDMTH bill by the same percentage, because the capacity charge and the fixed components do not scale with consumption.

What these systems do on a plant floor is narrower than the category name suggests. They read interval meter data, build a load profile per circuit, and issue scheduling or setpoint changes to the equipment that is actually under their control, which is rarely the process itself. The binding constraint is not the algorithm. It is whether the plant has interval metering below the revenue meter. Most plants do not, which makes the first capital decision a metering decision.

What the CFE bill is actually made of

CFE's briefing on the tarifas finales del suministro básico, issued in 2018 as the current scheme took effect, sets out what a Gran Demanda en Media Tensión Horaria account carries:

• a fixed monthly charge
• energy charges priced separately for base, intermedia and punta hours
• a capacity charge based on maximum demand coincident with the punta period
• a distribution charge
• transmission, CENACE operation, basic supplier operation and non-MEM ancillary service components

Read the third item again, because it is where control software earns or fails to earn its cost. The capacity charge is not set by average demand. It is set by the highest coincident demand the meter records in the punta period. A single compressor and chiller start that lands inside that window can set the charge for the month, and no amount of steady daytime efficiency elsewhere will unwind it.

The practical consequence is that two projects sold under the same heading have different economics. Trimming consumption across the day reduces energy charges only. Preventing one coincident start reduces the capacity charge, and that saving repeats every month the peak stays down. Anyone evaluating a platform should ask which of the two the vendor is quoting, and should read how GDMTH demand charges are actually set before agreeing a savings basis.

Where timing is worth money and where it is not

For a plant buying from CFE Suministro Básico, the value of moving load is fixed by the published tariff. The band differential and the capacity charge are the only two prices in play, and both are known in advance. Forecasting adds nothing to the price. It only helps you hit it.

For a plant that buys in the Mercado Eléctrico Mayorista, the picture is different, and the size of the difference is public. CENACE reports an average local marginal price of 878.23 pesos per MWh in the day ahead market across the Sistema Interconectado Nacional for the week of 10 to 16 March 2024. Across nodes in that same week the day ahead price ranged from minus 357.60 pesos per MWh to 12,524.42 pesos per MWh. The real time average was 782.09 pesos per MWh.

System load moved in the same week. CENACE records a maximum hourly purchase offer of 40,923.80 MWh on the Friday and a minimum of 28,163.18 MWh on the Sunday. Load that can be moved between those conditions is worth something to a market participant and worth almost nothing to a basic supply customer. Establish which one you are before you price a forecasting tool, and read forecasting industrial demand in practice alongside this.

What the software needs before it can do anything

The revenue meter tells you when billed demand was set. It does not tell you what set it. Attribution needs metering at circuit level on the largest loads, which on most industrial sites means the compressor room, the chiller plant, the main process feeders and, where they exist, the furnaces or ovens. That is a capital item with its own installation window and its own payback, and it comes before any software license. The groundwork is the same groundwork described in lowering operating costs with energy monitoring and IoT.

Once circuit level interval data exists, the findings that come first are not exotic. Compressor specific power drifting upward at constant delivered air points at leaks, filters or a failing stage. A chiller approach temperature that widens month over month points at fouled tubes. An oven or dryer that starts two hours before the first part arrives is a scheduling error that no algorithm is needed to see. These are the results of the first twelve months of data, and they are usually available before any model is trained.

One more prerequisite is organizational rather than technical. Someone has to own the data. A platform with no named owner produces alarms that nobody clears and a monthly report that nobody reads.

What an AI layer cannot control

On most industrial sites the production line is not under the energy management system. Line rate, batch sequence and furnace setpoints belong to production and to process engineering, and they are not going to be handed to an optimizer. Saying otherwise oversells the category.

What is genuinely controllable is the utility side. Compressed air scheduling and header pressure setpoint. Chilled water supply temperature and staging. HVAC in offices and non-process areas. Lighting. Non-critical pumping. Battery dispatch and vehicle charging where they exist. That list is smaller than a vendor deck implies and it is still worth working through, because those loads run whenever the plant is open and several of them are the ones that set the punta peak.

Monitoring is not Código de Red compliance

Energy platforms are often sold as covering regulatory obligations. In Mexico that claim needs testing. CRE issued the Código de Red in Resolution RES/151/2016, published in the DOF on 8 April 2016. It applies to load centers connected at media tensión, above 1 kV and up to 35 kV, and at alta tensión at 35 kV and above, regardless of contracted demand or supply scheme. Load centers connected before that date had until 9 April 2019 to comply.

The requirements are measurement requirements. Media tensión covers voltage variation, transient voltage, frequency variation, short circuit, protections, control systems, information exchange and current imbalance. Alta tensión adds power factor, current harmonic distortion and voltage flicker. CRE requires the studies to be made with Class A measurement instruments under NMX-J-610/4-3-ANCE or IEC 61000-4-30.

An energy monitoring platform is generally not built to that instrument class, because it exists to attribute cost rather than to certify power quality. The two data sets look similar on a screen and are not interchangeable. CRE does not authorize or certify inspectors for these studies, so the load center either performs them itself or engages a specialized firm. Treat the Código de Red work and the energy management work as two separate budget lines.

See what your interval data already shows

Mexico Energy Partners pulls the 15 minute interval history behind your CFE account and identifies where billed demand is being set and by which loads. The review reports how much of your bill sits in punta hour energy against how much sits in the capacity charge, and separates the savings available from scheduling alone from those that would need metering or controls investment. It states what that investment would cost. You supply twelve months of CFE invoices and the RPU for each service. If the plant has no interval metering, the report prices installing it before any software is discussed.

Sources

• Comisión Federal de Electricidad, Tarifas Finales del Suministro Básico, sesión informativa, 2018. Composition of the Gran Demanda en Media Tensión Horaria tariff, including the capacity charge based on maximum demand coincident with the punta period.
• CENACE, Reporte semanal del Mercado Eléctrico Mayorista, Sistema Interconectado Nacional, semana 11 de 2024, 10 to 16 March 2024. Average and extreme local marginal prices and hourly purchase offers.
• Comisión Reguladora de Energía, Resolution RES/151/2016, Código de Red, DOF, 8 April 2016. Scope by connection voltage, the measured parameters, the Class A instrument requirement and the 9 April 2019 compliance date.
• Comisión Reguladora de Energía, Guía sobre los requerimientos técnicos del Código de Red aplicables a Centros de Carga, gob.mx. Confirmation that CRE does not authorize or certify the parties who perform these studies.