Hydrotreating margin usually erodes through small cushions in temperature, hydrogen and throughput, long before anyone calls it a problem. According to the U.S. Energy Information Administration, merchant suppliers met 70% of U.S. refinery hydrogen demand in 2022, so every extra cushion of hydrogen carries a direct cost.
Internal ultra-low sulfur diesel (ULSD) targets often sit below the formal limit to cover analytical variability. That pushes operators toward conservative settings for weighted average bed temperature (WABT), hydrogen-to-oil ratio and throughput. Conventional advanced process control (APC) struggles to coordinate sulfur removal, catalyst life and hydrogen consumption at the same time, and hydrotreating optimization depends on handling all three together.
How conservative margins compound cost
WABT
Reactor WABT is the most influential variable for diesel sulfur. A higher WABT removes more sulfur, and it also increases cracking, reduces diesel yield, accelerates catalyst aging and raises energy use. A unit running 5°F above the WABT it needs may give up yield and catalyst life for sulfur margin the current feed did not require.
Hydrogen
Many units blend feeds and treat them at one conservatively high severity, which over-treats lighter fractions that are already close to spec. Feed segregation and severity matching can cut hydrogen consumption. The coordination it needs across feed tanks, reactors and scheduling constraints is hard to sustain by hand across shifts. When merchant hydrogen prices spike, the cost of that conservatism is harder to absorb.
Energy and throughput
Hydrotreaters are significant energy consumers in a refinery. Poor heat integration, conservative furnace firing or excess quench gas keep that burden high. Combined with hydrogen cost and throughput held in reserve to protect end-of-run margin, conservative operation adds cost in three places at once.
Regulatory pressure
Regulation keeps those cushions in place. Tighter sulfur specifications raise the stakes for units that already run conservatively. When a hydrotreater misses internal sulfur targets, the cost lands directly in operations through off-spec product disposition, credit exposure or lost blending flexibility.
The penalty for missing spec makes conservative operation feel rational in the moment, even though the cost adds up over a full catalyst cycle.
Why end-of-run pressure builds faster than expected
Catalyst deactivation turns margin pressure into a cycle-length problem. Activity loss is continuous and nonlinear, and it often accelerates near end-of-run even when projected catalyst life looked manageable earlier in the cycle. A unit operating near its temperature limit can still lose run length when the end-of-run temperature rise steepens faster than expected.
Above-baseline deactivation usually traces to coke formation, metals deposition or poisoning by nitrogen compounds. Polyaromatic content and light cycle oil blending rates correlate strongly with faster aging, and low hydrogen-to-oil ratios can multiply the effect. Feed variability adds to this when heavier or more contaminated material pushes operators to raise severity to protect product specs. Metals add a separate constraint, because catalyst-bed fouling can raise pressure drop and cause flow maldistribution even while activity remains acceptable.
End-of-run is rarely a temperature issue alone. Pressure drop, product sulfur, color and throughput can all become limiting together, which makes simple extrapolation unreliable. In integrated refineries those limits also have to line up with turnaround schedules and the performance of adjacent units. A hydrotreater that reaches its temperature ceiling two months early can force a refinery-wide replanning exercise.
Most refineries run several hydrotreaters feeding different product pools. A cycle-length surprise on one unit ripples into blending, scheduling and hydrogen allocation decisions across the plant.
Where independent control loops fall short
Conventional control manages hydrotreating constraints through loops that are largely separate, and the APC layer above them relies on linear models identified around one set of operating conditions. Neither was built for a unit where sulfur specification, hydrogen consumption, catalyst life, throughput and temperature all interact nonlinearly.
Feed changes
The gap appears first when feed composition shifts. Conventional APC often corrects after a quality measurement has already moved, while hydrogen partial pressure moves quickly with feed changes, catalyst aging and supply. Operators respond by padding hydrogen ratios and accepting higher energy cost to avoid catalyst damage. That leaves a persistent gap between what the unit could run and what it does run.
Bed temperatures
Temperature control has a parallel problem. Exothermic hydrotreating reactions can create local hot spots, and even moderate WABT deviations affect sulfur and nitrogen removal. When a hot spot develops in one bed, APC may adjust quench to that bed without accounting for the effect on downstream beds or overall hydrogen consumption. The adjustment is local, and its consequences propagate downstream.
Catalyst aging
Without a forward projection of aging, operators trade margin erosion today against reliability risk months from now. By the time the trajectory is clear, the options have narrowed. Many hydrotreaters spend much of their cycle running more conservatively than the catalyst condition requires.
What coordinated optimization changes
The features that make hydrotreating margin hard to manage by hand are the same ones that suit a learned model: nonlinear behavior, competing constraints, and decisions that unfold on different time scales.
A model trained on plant history can learn the relationships between feed composition, WABT, hydrogen partial pressure and product quality that linear models do not capture well. It coordinates the full constraint set, so temperature, hydrogen flow and throughput move together as feed quality shifts.
It also brings the time scales together. Temperature moves play out over minutes, hydrogen dynamics over hours, and catalyst deactivation unfolds over months. A coordinated model keeps all three horizons in view, so today's operating point reflects where the catalyst is heading as well as where it is. A unit can then run closer to its actual constraint early in the cycle and back off gradually as the catalyst ages, instead of holding a fixed cushion from day one.
Coordinating across the hydrogen network
When multiple units share a hydrogen network, purity and flow decisions in one part of the system shift cost somewhere else, and static tools often miss that interaction. A model that sees the whole network can send hydrogen where it creates the most value, instead of defaulting to equal allocation or manual priority rules, while keeping product quality constraints in view. Where the hydrogen balance is already tight, small gains in allocation efficiency can lower overall energy cost. For the hydrocracker side of the same balance, see hydrocracker control and hydrocracker yields.
No model substitutes for the judgment that comes from years at the board. Routine WABT, hydrogen and throughput rebalancing happens continuously, though, and no manual process sustains it consistently across every shift. The model handles the routine adjustments, and the situations that need operator experience still get it.
How trust builds
Plants that build lasting trust with optimization usually start in advisory mode. The model recommends setpoint changes, operators test them against their own experience, and confidence builds step by step.
On a hydrotreater, advisory mode shows how WABT targets, hydrogen ratios and throughput limits interact across feed changes and catalyst aging. Because the model tracks deactivation alongside daily decisions, it can flag when a cushion is no longer needed or when conditions ahead call for pulling back. Planning, operations and engineering work from the same model of unit behavior, so decisions about cycle timing and turnaround coordination rest on current conditions. Newer operators can see the reasoning behind experienced decisions, and veterans can see their intuition checked against actual unit performance.
Plants move from advisory mode to closed-loop optimization when operating goals and trust support it. Operators set the envelope the controller may use and can take over from the board.
How the Imubit Platform applies
Imubit applies Map, Model and Control to hydrotreating as part of its work in refining. Map traces conservative cushions in WABT, hydrogen and throughput to the recurring decisions behind them. Model learns how feed, catalyst condition, hydrogen partial pressure and product sulfur interact from the unit's own history, using Deep Learning Process Control® (DLPC®). Control takes those decisions to closed loop, writing setpoints to the existing APC and DCS, inside limits set by plant engineers, so the cushion can narrow when feed and catalyst allow and widen again when they do not.
