The Models Behind Thermal Generation Are Running Towards Obsolesce

A combined-cycle plant commissioned fifteen years ago was designed, financed, and modeled to run flat out as baseload. It does not run that way anymore. To balance a grid filling with intermittent renewables it now starts, stops, and ramps constantly, and at the same time it is being ordered to stay available for a demand curve climbing faster than it has in years. The North American Electric Reliability Corporation projects winter peak demand up more than twenty gigawatts on last year, driven substantially by AI data centers, and since January 2026 the US Department of Energy has issued a run of emergency orders under Section 202(c) compelling generators to remain online and dispatch to hold reliability. The plant has been given a new job. 

It is running that new job on the models of the old one. The heat-rate curve, the maintenance interval, and the emissions profile the operator plans against were all calibrated for steady baseload, and none of them describes a plant that now cycles. The result is an operator held accountable for the efficiency, the reliability, and the component life of an asset whose own reference models quietly went out of date.

Every number an operator uses to run a thermal plant assumes it runs the way it was designed to. The OEM performance curve states a heat rate at design load and steady conditions. The maintenance schedule counts operating hours toward an overhaul interval built for continuous duty. The emissions rate is a figure measured at stable output. Under cycling, all three drift from reality at once and in different directions, because starts, stops, and ramps stress a plant in ways continuous running never did. The operator is not short of data; the plant is instrumented to the teeth. What is missing is a model of the plant as it behaves now, rather than as it was specified to behave then.

The first casualty is efficiency. Flexible operation has a substantial negative effect on heat rate, because constant swings in temperature and pressure pull the plant away from its efficient operating point and hold it there through the ramps and part-load hours that cycling forces (NREL; EPRI). Turbine efficiency that sits at 88 to 91 percent fresh out of an overhaul drifts into the low eighties as the cycles accumulate, and every point of heat rate lost is fuel burned for no output and carbon emitted for no megawatt. Because the design curve never modeled this duty, the loss is invisible on paper. The operator sees a fuel bill higher than the model says it should be and cannot tell how much is degradation, how much is duty, and how much is ambient, which is exactly why it cannot be recovered.

The second casualty is the asset itself. Cycling consumes component life on a completely different schedule than baseload running, because thermal fatigue accumulates with every start and stop, and EPRI's review of four decades of failure data puts boiler tube leaks as the single most common cause of failure in conventional steam plants, a mode that cycling accelerates. The overhaul and replacement schedule, though, still counts operating hours against intervals built for continuous duty, so it mis-times the one decision that matters most economically: when to take a unit offline and replace the parts that are actually wearing. Plan the overhaul too early and capital is wasted; too late and the forced outage arrives at the worst possible moment, which under the current reliability orders is also the most exposed one. This is predictive asset replacement, but the question is not whether a part will fail. It is how much life the plant's real duty has already spent, which an hours-based schedule cannot see. 

The pressure is not coming from where operators spent a decade expecting it. Federal carbon regulation is loosening, not tightening: the Environmental Protection Agency sent a final rule to repeal the greenhouse-gas standards for fossil plants for review in May 2026, having already repealed the updated coal mercury standards earlier in the year. The tightening is on reliability. The Section 202(c) orders make availability a legal requirement rather than a commercial preference, NERC's cold-weather standards require generator owners to report winterization status by June 1, 2026 and hold them to performance obligations, and the AI-driven demand surge means the grid is leaning on exactly the aging, cycled plants least able to promise it. An operator is now obligated to guarantee the availability and performance of assets whose true condition its own models no longer describe, and the cost of guessing wrong has moved from a line on a spreadsheet to a compliance finding. 

Closing that gap does not take more sensors. It takes a model of the plant as it actually runs today, learned from the plant's own signal under its real cycled duty rather than inherited from a design sheet. A self-learning system trained on the unit's own starts, ramps, temperatures, and outputs reports the true heat rate the plant is achieving now, separates genuine degradation from duty and ambient so the recoverable efficiency can actually be recovered, tracks how much component life each cycle has really consumed so overhauls and replacements are timed to wear rather than to a calendar, and states the plant's real emissions and real available capacity when a reliability order lands. CodeNinja builds these systems inside the operator's own environment and transfers them in full at close, weights, training data, and pipelines included, so the model of the plant belongs to the operator that runs it. The firm whose performance curve the plant was bought on cannot supply that model, because its curve is a product sold across the fleet, not a mirror of your unit. 

The thermal fleet is not being retired on the schedule anyone drew up. It is being kept online, cycled harder, and leaned on to carry a grid through a demand surge it was never sized for, and the operators who come through that with margin and compliance intact will be the ones who run their plants against what those plants are actually doing rather than against the design they have outgrown. The OEM curve, the hours-based schedule, and the nameplate emissions rate describe a plant that retired years ago in everything but name. A model of the real plant, owned by the operator, is how efficiency is recovered, how life is spent deliberately, and how availability is proven to a regulator that is no longer asking politely. 

CodeNinja runs a structured Discovery Session for thermal generation operators and fleet owners. It reviews the heat-rate, maintenance, and emissions models your plants run on today, shows where cycling has pulled the real plant away from them, and maps what an owned, self-learning model of the plant as it actually operates looks like across efficiency, component life, and reliability. Start that conversation at https://codeninjaconsulting.com/contact.

• North American Electric Reliability Corporation (NERC). Winter reliability assessment (winter peak demand growth; cold-weather winterization reporting due 1 June 2026). 
• U.S. Department of Energy. Section 202(c) emergency orders, 2026. 
• U.S. Environmental Protection Agency. Repeal of greenhouse-gas emissions standards for fossil-fuel-fired power plants (final rule submitted for review, 2026); repeal of updated coal mercury standards, 2026. 
• National Renewable Energy Laboratory (NREL), Power Plant Cycling Costs; EPRI analysis of flexible-operation impacts on heat rate and component degradation.