Catalytic Reforming Optimization: Balancing Octane, Hydrogen, and Catalyst Economics

Every barrel of naphtha that enters the catalytic reformer carries a decision embedded in its chemistry: how much octane to push for, how much hydrogen to produce, and how fast to spend the catalyst doing it. That decision cascades through downstream operations, from gasoline blending targets to hydrotreater hydrogen supply.

With US Gulf Coast refining margins falling more than 50% from recent peaks, the consequences of getting that balance wrong compound quickly. Reformer optimization has shifted from a continuous improvement project to a core profitability requirement, and the tools most refineries rely on haven’t kept pace with the complexity involved.

TL;DR: Catalytic Reforming Optimization for Octane, Hydrogen, and Margin Recovery

Catalytic reforming presents a high-impact optimization opportunity in the refinery, but traditional control approaches struggle with its constantly shifting dynamics.

Why Reformer Optimization Outgrows Linear Control

• Traditional APC relies on linear approximations that can’t capture the temperature-dependent reaction kinetics and catalyst deactivation inherent to reformers.
• When models degrade, operators disable APC entirely and revert to conservative manual control.

How AI Optimization Adapts to What Changes Every Day

• Adaptive models track catalyst aging and feed variability to update severity recommendations continuously, rather than waiting for quarterly retuning.
• The economic impact comes from avoiding reforming’s two most common failure modes: running too conservatively when constraints support more octane, and pushing late-cycle catalyst in ways that accelerate deactivation.

The sections below explore why reformer optimization demands a different approach and what that looks like in practice.

The Severity-Octane Balancing Act and Its Margin Consequences

Reformer severity sits at the center of a daily tension. Push severity and the unit typically delivers higher-octane reformate while accelerating catalyst deactivation. Back off severity and catalyst life improves, but the refinery may lose octane barrels it needs to meet blending specifications.

That trade-off is rarely limited to octane alone. A severity move can also change reformate yield, hydrogen make, and the load on downstream separation and compression. In practice, operators often feel it as a set of coupled constraints: heater duty limits, reactor temperature approach limits, recycle gas compressor margin, separator conditions, and the need to stay inside product quality specs.

Each octane point carries real economic weight. A single RON increase can translate into several dollars per barrel of margin improvement depending on gasoline market conditions. The cost side is just as real: catalyst replacement and the associated outage can be a multi-million-dollar event, so the timing of changeouts or regenerations becomes a high-stakes economic decision.

Feed Shifts and the Moving Target

Feedstock variability compounds the problem because severity isn’t one knob. Naphtha endpoint temperature, sulfur content, and paraffin-to-naphthene ratios all move the octane and hydrogen response curve. A feed shift that brings more paraffins may call for higher temperatures to reach the same octane, while a shift toward more naphthenes can make the same temperature strategy overshoot aromatics and hydrogen.

And the target keeps moving. “Optimal severity” looks different at start of run than it does late in the cycle, as declining catalyst activity demands higher temperatures and tighter constraint management to reach the same targets.

When uncertainty about feed, catalyst condition, or downstream requirements rises, the unit tends to drift toward a conservative posture: extra temperature, extra recycle, extra cushion from a hydrogen pinch. Those cushions protect the unit, but they increase energy consumption and accumulate margin loss across weeks and months.

Why Reformer Optimization Outgrows Linear Control

Traditional advanced process control (APC) assumes relationships between variables remain constant. In a reformer, they don’t. A severity change that produces a predictable octane response with fresh catalyst can produce a different response months later, even with the same setpoints.

The reaction kinetics are nonlinear, temperature-dependent, and coupled across multiple reactors in series, so every setpoint adjustment interacts with a process that has already shifted since the model was last tuned.

Reformers have interacting mechanisms that shift at different speeds. Temperature changes drive immediate conversion shifts, while chloride balance, coke deposition, and feed composition changes reshape the response over weeks. A linear model may look acceptable in a short validation window and still drift when those slower effects take hold.

What Happens Between Tuning Cycles

Between tuning cycles, multiple sources of drift compound. Operators often infer octane from analyzers, lab samples, and estimators whose bias shifts with feed and operating point. Lab RON results arrive on a delay, so the controller may be tuned against a proxy that has already moved.

Meanwhile, models require quarterly or semi-annual retuning by specialized engineers, and performance degrades steadily in the intervals between. The unit can appear “under control” while quietly giving away margin.

The most telling consequence is behavioral. When a controller becomes unreliable, operators frequently disable APC and revert to manual moves with wider safety margins. That response is rational: a controller that occasionally pushes a constraint the wrong way on a night shift is a liability the plant can’t afford. Linear models weren’t built for a process where the cost of one wrong move can be measured in lost cycle length or lost blending flexibility.

How AI Optimization Adapts to What Changes Every Day

AI optimization earns its place in reforming by doing what linear controllers can’t: adapting as conditions change rather than assuming a static snapshot holds. AI models trained on plant data learn how the process behaves across catalyst age and feed variability, then update recommendations accordingly.

The distinction matters most in catalyst management. As coke deposits build and activity declines, the operating strategy that made sense early in the cycle can become overly expensive later. An adaptive model can learn how the catalyst has behaved over previous cycles and update severity recommendations as it ages, instead of waiting for manual retuning.

In operations terms, the model can incorporate signals that experienced operators already watch, but at a resolution and consistency that’s hard to sustain manually. Reactor temperature profiles and inter-reactor temperature rises, recycle hydrogen rate and purity, separator stability, and unit response to small temperature nudges all carry information about where the unit sits on its catalyst curve.

When those signals move together, an adaptive optimizer can distinguish “feed got heavier” from “activity is slipping,” which leads to different severity and hydrogen strategies.

Where the Margin Recovery Shows Up

The economic impact tends to come from avoiding two common failure modes. One is running too conservatively for too long because the unit “feels late-cycle,” even when feed and constraints would support more octane. The other is pushing a late-cycle catalyst in a way that looks fine for a few days but accelerates deactivation and pulls the end-of-run forward.

Tighter, more repeatable performance across day-to-day variability reduces the need for wide operating cushions and recovers margin that compounds over full catalyst cycles. When planning, operations, and commercial teams coordinate around a shared view of the unit, refineries have captured margin improvements of $0.50 to $1.00 per barrel through better-aligned optimization decisions.

Building Operator Trust in Stages

Implementations that succeed start in advisory mode, where operators evaluate AI severity recommendations against their own read on compressor behavior, heater firing stability, and downstream constraints before acting. Over time, if the optimizer repeatedly recommends smaller, earlier severity moves rather than large step changes, operators can see how that reduces overshoot and improves constraint stability.

That feedback loop makes the system a transparent tool operators can trust. It also captures what good operation looks like, so the knowledge that a thirty-year veteran carries doesn’t walk out the door when they retire. Instead, it stays embedded in the model, available on every shift as experienced operators hand off complex units to newer teams.

Connecting Reformer Decisions to Refinery-Wide Economics

Reformer optimization in isolation misses the point. Severity adjustments change hydrogen production, which changes hydrotreater capacity, which changes the crude slate the refinery can process. Octane output affects gasoline blending economics, and aromatics content determines whether reformate is worth more as blendstock or as petrochemical feedstock.

How Severity Moves Cascade Through the Site

These interactions show up as practical trade-offs that operations teams manage daily. When the hydrogen system tightens, a “simple” reformer severity increase can constrain a hydrotreater, which can push the site toward a different crude or force a throughput cut.

The reverse happens too: when hydrotreating demand drops, the reformer may be able to run a severity strategy that favors octane and reformate yield without risking a hydrogen pinch. Most refineries manage these interactions through LP planning models updated periodically, with planning setting targets based on steady-state assumptions that can lag reality by weeks.

The Gap Between Planning Models and Current Catalyst Condition

The planning-operations gap is widest during late-run deactivation, unusual feed slates, or post-maintenance restarts. A plan might assume an octane target is “available,” while operations sees that current catalyst condition requires a severity move that puts heater duty or recycle compression uncomfortably close to limits. Refineries may schedule regeneration cycles based on calendar intervals, while real-time performance data suggests the unit is deactivating faster or slower than expected.

When all three functions reference a single model of how the reformer actually behaves today, the dynamic changes. Planning teams gain visibility into how current catalyst condition affects achievable octane targets. Operations understand the refinery economics reasoning behind severity recommendations, and maintenance can time regeneration decisions against actual performance degradation rather than conservative schedules.

Cross-functional transparency is where the compounding value lives. The reformer doesn’t operate in a vacuum, and its optimization framework shouldn’t either. When hydrogen production, energy consumption, octane targets, and catalyst lifecycle all feed into a shared model, operations teams capture margin that unit-level approaches leave behind.

Recovering Margin Across the Full Catalyst Cycle

For refinery operations leaders managing the tension between octane demand, hydrogen supply, and catalyst economics, Imubit’s Closed Loop AI Optimization solution offers a path forward. The platform learns from each reformer’s unique operating history, adapts to catalyst aging and feed quality shifts, and writes optimal setpoints in real time through the existing distributed control system (DCS).

Plants can start in advisory mode, where operators evaluate AI recommendations alongside their own expertise, and progress toward closed loop operation as confidence builds. The result is reformer optimization that responds to what’s actually happening in the unit today rather than what a static model assumed months ago.

Get a Plant Assessment to discover how AI optimization can recover margin from your catalytic reformers by aligning severity, catalyst management, and hydrogen production with real-time refinery economics.

Frequently Asked Questions

How does catalyst deactivation affect the accuracy of reformer optimization models?

Catalyst deactivation changes the reformer’s response to operating variables over time, so static models drift from actual behavior within weeks of calibration. Traditional APC typically needs manual retuning to correct that drift. Between interventions, the controller runs on stale assumptions. Approaches built for continuous process control can adapt to changing activity and keep recommendations aligned with current unit behavior rather than last quarter’s tuning.

Can AI optimization work alongside existing APC systems on a catalytic reformer?

AI optimization typically integrates with existing APC and control system infrastructure rather than replacing it. The AI layer operates above existing controls and adjusts setpoints that APC then executes at the regulatory level. This architecture avoids decommissioning current control systems while adding recommendations that account for nonlinear dynamics and catalyst condition that APC alone can’t model. The integration approach also aligns with how many plants extend advanced process control without disrupting operator workflows.

What metrics should refineries track to measure catalytic reformer optimization success?

The most useful metrics tie unit performance to site economics: reformate yield at target octane, hydrogen production per barrel of feed, energy consumption per barrel of reformate, and catalyst cycle length versus historical baseline. Tracking shift-to-shift variability in those KPIs also shows whether improvements hold across operating teams. Plants that connect these measures to planning targets usually see faster alignment on constraints and clearer prioritization using shared operational efficiency metrics rather than debating unit-level heuristics.