Nuclear power plants were never designed to do what the 21st-century electricity grid increasingly demands of them: follow the sun and chase the wind. Built in an era of baseload dominance — where the job was simply to run flat out, all the time — pressurized water reactors (PWRs) are instead being asked to ramp up and down daily, filling in the gaps left by solar and wind generation. That's a harder problem than it sounds, and for decades it's been one of the quiet bottlenecks holding back a cleaner grid.
A new system developed by engineers at Framatome and described in a paper by Dupré and Grossetête (2026) proposes an elegant solution. They call it OAPS — the Optimal Axial Power System — and its central metaphor tells you almost everything: it is a GPS navigator for nuclear operators. Where a GPS continuously recalculates your route based on where you actually are, OAPS continuously recalculates control recommendations based on the latest real measurements from inside the reactor. The result, the authors argue, is a system that can determine the fastest safe ramp rate, suppress dangerous power oscillations before they grow, and minimize the volume of radioactive water that gets discharged during a maneuver — all at once, in real time.
The paper, presented at the International Congress on Advances in Nuclear Power Plants (ICAPP 2025), showcases three new advanced strategies on an intermediate-complexity PWR simulator developed by Framatome, building on prior validation work done on both Framatome's own full-scope PWR simulator and EDF's full-scope N4 simulator. It represents one of the most operationally concrete applications of advanced control theory to civilian nuclear power yet published.
The Science
To understand why OAPS matters, you need to understand xenon. Specifically, xenon-135 — a radioactive gas produced as a byproduct of uranium fission inside a reactor core. Xenon-135 is a voracious absorber of neutrons, meaning it acts as a natural brake on the chain reaction. When reactor power drops, xenon builds up (because the neutron flux that would normally destroy it is reduced). When power rises again, xenon burns off. This creates a deeply non-linear feedback loop with a time constant of many hours — a phenomenon so significant it contributed to the 1986 Chernobyl disaster, where operators underestimated xenon's role in the reactor's behavior during a test.
For modern, well-designed PWRs, xenon doesn't pose that kind of existential risk. But it does create a fiendishly complex control problem. Power changes propagate unevenly through the core, setting up what engineers call "axial power oscillations" — waves of high and low neutron activity moving up and down the length of the reactor, driven by xenon redistribution. Left unchecked, these oscillations can grow and force operators to limit their power maneuvers, or to inject large quantities of boron (a neutron absorber dissolved in the coolant water) to stabilize the core. Boron injection is effective, but it generates diluted radioactive water that must be processed and managed. Every liter is a small operational cost; over a plant's lifetime, those liters add up.
The OAPS system attacks all of this simultaneously using a technique called model predictive control (MPC). MPC — well-established in chemical processing and aerospace but relatively novel in nuclear operations — works by maintaining a mathematical model of the system's future behavior and continuously optimizing a sequence of control actions over a rolling time horizon. At each step, the controller asks: "Given where the plant is right now, what sequence of actions over the next N hours minimizes a cost function — say, time to reach target power, or total boron used — while staying within all safety limits?" It executes the first action, gets new measurements, and repeats. This is exactly how a GPS navigator works: it doesn't plan your entire trip once and stick to it. It re-optimizes every few seconds based on where you are.
The physics model inside OAPS captures the core neutronics (how neutrons behave), xenon and iodine dynamics (iodine-135 decays into xenon-135, making the chain one step more complex), and thermal-hydraulics — how heat moves through the fuel and coolant. The control variables the system recommends include dilution and boration flowrates (how much boron to add or remove from the coolant), turbine power setpoints (how much steam to draw from the reactor), and the variation rates at which those changes should occur. These are the exact quantities a human operator manipulates; OAPS doesn't replace the operator but advises them, in real time, with optimized numbers.
What They Found
The paper demonstrates three distinct operational scenarios, each illustrating a different facet of what OAPS can do.
Fastest feasible power variation. One of the persistent frustrations of flexible nuclear operation is that operators don't always know how fast they can safely ramp power. The conservative answer is slow — follow a cautious predetermined ramp rate and don't push the physics. OAPS takes a different approach: it solves, in real time, for the maximum ramp rate that keeps the plant inside all operational limits simultaneously. The system accounts for the current xenon distribution in the core, the current boron concentration, and the thermal state of the fuel — and returns the fastest path that is provably safe given those conditions. The practical implication is significant: on days when the reactor starts a maneuver from a favorable xenon state, OAPS may find a path substantially faster than the standard conservative rate. On days when the xenon distribution is less favorable, it constrains the rate accordingly, preventing operators from inadvertently pushing the plant toward a limit they can't see.
Accelerated cancellation of xenon oscillations. When axial power oscillations begin — as they inevitably do during or after a significant load-follow maneuver — the conventional response is a slow, careful application of control rods and boron adjustments. OAPS demonstrates an accelerated suppression strategy: by modeling the xenon wave's future trajectory, the system identifies the control actions that will dampen it in the shortest time, rather than simply riding it out. The authors show this working on the Framatome intermediate-complexity simulator, with the oscillation being brought under control noticeably faster than with conventional methods. Faster oscillation suppression means the plant spends less time in a constrained state and returns to full operational flexibility sooner.
Minimization of water and boron effluents. This is perhaps the most quietly significant result. Every time a PWR changes power significantly, some volume of borated water is exchanged between the primary coolant circuit and the plant's liquid waste system. This "effluent" is low-level radioactive waste — not dangerous in small quantities, but costly to process, store, and ultimately dispose of. By optimizing the boration and dilution strategy, OAPS finds maneuver paths that achieve the same power change while minimizing the total volume of water exchanged. The system treats effluent volume as a term in its cost function: not just "get to the target power" but "get to the target power using as little boron water as possible." The authors demonstrate this on the simulator, showing meaningful reductions compared to conventional operator practice.
Why This Changes Things
The bigger picture here is about what flexible nuclear power could mean for the clean energy transition. Wind and solar are now the cheapest sources of new electricity generation in most of the world, and their share of grids is rising fast. But they're variable — output depends on weather, not demand. The standard solution is a mix of storage, interconnection, and "dispatchable" generation that can ramp up and down on demand. Natural gas has historically filled that dispatchable role. But if the world is serious about decarbonizing electricity, gas has to go — and something else has to pick up the flexibility role.
Nuclear is the obvious candidate in many grid configurations. It's carbon-free, reliable, and the fuel energy density means a single plant can generate enormous amounts of power. But its historical inflexibility — the tendency to run at constant output — has been both a technical and cultural constraint. Technically, xenon physics makes aggressive load-following genuinely difficult. Culturally, nuclear operators are trained to prioritize stability above all else, and variable operation feels risky. France, whose grid is over 70% nuclear, has spent decades developing load-following protocols for its PWR fleet — but even French operators face real limits on how fast and how deeply they can vary output.
OAPS directly addresses both the technical and cultural constraints. On the technical side, it makes the optimal path explicit and computable, removing guesswork from maneuver planning. On the cultural side, it gives operators something they currently lack: a continuously updated, quantitative answer to "what can I safely do right now?" A GPS doesn't make driving less safe — it makes drivers more confident navigating routes they'd otherwise avoid. The analogy is apt.
The system's prior validation on EDF's full-scope N4 simulator is worth noting. The N4 is one of EDF's most powerful reactor designs, a 1,450 MWe PWR, and full-scope simulators replicate the actual control room environment with high fidelity. Validation there means the system has been tested against physics models and operator interfaces that closely mirror the real thing. The new results on Framatome's intermediate-complexity simulator extend that work into more operationally nuanced territory.
There's also a cost angle. Boron effluent minimization reduces waste processing costs. Faster power variations reduce the time a plant spends at partial power during maneuvers — and a reactor not at full power is a reactor not selling electricity. The combination of operational efficiency and waste reduction makes OAPS an economically attractive proposition, not just a safety-engineering achievement.
It's worth being clear about what MPC brings to this problem that simpler control methods don't. Conventional PWR control relies on pre-computed operational charts — maps showing which combinations of axial offset and power are permissible — and rule-based guidance for how to adjust boron and rod positions. These charts are conservative by design: they're drawn to be safe under a wide range of conditions, not optimized for any particular condition. MPC, by contrast, continuously solves an optimization problem conditioned on the current actual state of the plant. It is, in a precise sense, doing the best possible job given where the plant is right now, rather than the best average job over all possible conditions. That's the difference between a GPS and a road atlas.
What's Next
The honest caveat is that demonstration on a simulator, even a high-fidelity one, is not deployment in a real plant. The gap between "validated on full-scope simulator" and "approved for operational use in a licensed nuclear power plant" is substantial. Nuclear regulators — the Nuclear Regulatory Commission in the US, the Autorité de Sûreté Nucléaire in France — require extensive safety cases, and an advisory system that gives operators real-time recommendations would need to demonstrate not just that its recommendations are usually good, but that its failure modes are well-characterized and bounded. What happens if the plant's measurement system gives OAPS a bad reading? What happens if the model diverges from actual plant behavior? These are not insurmountable problems — aviation's fly-by-wire systems and aircraft autopilots have navigated similar questions — but they require serious engineering work.
The paper also notes that the three new strategies were demonstrated on an intermediate-complexity simulator, a step below the full-scope validation already achieved for the conventional axial offset control strategy. Full-scope validation of the new strategies — fastest ramp rates, oscillation suppression, effluent minimization — presumably remains ahead.
There's also the question of integration with grid operators. A system that can tell a nuclear plant how fast it can ramp is only useful if the grid operator knows what flexibility is available, and can request it in a timely way. This connects OAPS to the broader challenge of nuclear plants participating in modern electricity markets, which increasingly dispatch generation in 5-minute intervals or faster. The technical capability to ramp faster is a prerequisite; the market and communication infrastructure to use that capability is a separate piece of work.
None of these are reasons for pessimism. They're the expected shape of a promising technology moving from research toward deployment. What Dupré and Grossetête (2026) have demonstrated is that the core technical problem — how to optimally navigate the complex, xenon-dominated physics of a PWR during a power maneuver — is genuinely solvable with modern control theory, and solvable in real time. That's the finding that matters.
The clean energy transition has many hard problems. Flexible, dispatchable, zero-carbon generation is one of the hardest. If nuclear plants can learn to follow the wind and the sun — not grudgingly, not at the cost of safety margins, but confidently and optimally — the shape of the future grid looks meaningfully different. OAPS, for all its technical specificity, is an early piece of that picture.