Hydrocracker Optimization with AI: Capturing the Margin that Traditional Controls Miss

Feedstock composition shifts mid-run. Catalyst activity continues its slow decline. And the control system keeps writing the same setpoints it calculated yesterday. Hydrocracker operators know this frustration well: the gap between where the unit could run and where conservative controls keep it widens with every feed change the APC wasn’t tuned to handle.

That gap represents real money. BCG reports that downstream earnings for integrated oil companies dropped about 50% in 2024 versus 2023, intensifying pressure on unit-level performance. Value chain optimization research from McKinsey suggests North American refineries could realize $0.50 to $1.00 per barrel improvement through better optimization across the enterprise. Hydrocrackers, with their complex interactions between conversion, selectivity, hydrogen cost, and catalyst life, represent one of the highest-leverage units for capturing a share of this value.

The economics of hydrocracker optimization center on conversion, middle-distillate yield, and the ability to respond to changing conditions without sacrificing margins to conservative setpoints. These factors interact in complex, nonlinear ways that traditional controls struggle to navigate.

TL;DR: How AI Captures Hidden Hydrocracker Margin

AI optimization addresses fundamental limitations in advanced process control (APC), enabling refineries to capture margin that conventional systems leave on the table.

Why Traditional APC Falls Short

• Linear models from step tests cannot capture nonlinear dynamics that shift with feed and catalyst conditions
• Reidentification cycles leave days or weeks of suboptimal operation after process changes
• Conservative WABT setpoints sacrifice conversion to maintain safety buffers

Where AI Captures Value

• Temperature profiles adapt to catalyst deactivation while respecting metallurgical limits
• Quench distribution shifts with feed composition to protect catalyst and maximize conversion
• Hydrogen-to-oil ratios balance cost against product specifications in real time

Here’s how these capabilities translate into recovered margin.

Why Traditional APC Falls Short on Hydrocrackers

Operators understand their units. Traditional controls simply cannot adapt fast enough when conditions change.

Conventional APC relies on linear gain matrices derived from step tests conducted during commissioning. These models assume process behavior remains constant, but hydrocracker dynamics shift continuously. Feed composition changes reaction kinetics in ways that linear models cannot predict. Catalyst activity declines over time, altering the relationship between temperature and conversion that the original model captured. Seasonal product demand shifts add another layer of variability that fixed setpoints cannot accommodate.

This creates a compounding problem. When operators push toward constraint boundaries to maximize conversion, traditional controls lack the confidence to operate there safely. Conservative weighted average bed temperature (WABT) setpoints become the norm, keeping units safely below metallurgical limits but sacrificing conversion that drops directly to the bottom line. Every degree of severity left unused represents yield that could have been captured. The margin left on the table accumulates shift after shift.

Operating constraints remain non-negotiable: maximum temperature rise limits protect catalyst beds and reactor metallurgy, pressure boundaries keep hydrogen in solution, and product specifications for diesel sulfur and jet aromatics cannot be treated as soft targets. These boundaries must exist. The problem is that static control models cannot navigate them dynamically. When feed quality shifts or catalyst ages past a threshold, reidentification through step testing can take days or weeks while the unit operates suboptimally. During that window, operators face a choice between running conservatively or accepting risk that existing models no longer accurately capture.

Where AI Optimization Captures Value

The opportunity lies in coordinating across the variables that drive hydrocracker economics simultaneously, rather than optimizing individual loops in isolation. All optimization operates within existing safety interlocks, alarm limits, and design constraints.

Temperature profile management represents the largest opportunity. The optimal profile depends on feed characteristics, catalyst age, and desired product slate. AI models learn the relationship between bed temperatures, conversion, and selectivity across thousands of operating scenarios, then adjust profiles continuously as conditions change. When feed nitrogen content increases, the model recognizes the impact on catalyst activity and adjusts severity accordingly. When catalyst deactivation accelerates mid-run, profiles adapt without waiting for the next step test cycle. Operators can validate these recommendations in advisory mode before granting authority for direct setpoint changes. This approach enables operation closer to optimal severity without exceeding metallurgical or catalyst limits.

Quench distribution directly impacts both conversion and catalyst life. Traditional controls distribute quench based on fixed ratios or simple temperature targets. But optimal distribution shifts with feed composition and catalyst deactivation. Heavier feeds generate more heat in the first reactor, requiring different quench allocation than lighter feeds that distribute heat more evenly across beds. When models adapt to these changes, they maintain temperature profiles that maximize conversion without creating hot spots that accelerate catalyst degradation. The result is both improved conversion and extended catalyst cycle length.

Hydrogen management balances competing objectives. Higher hydrogen partial pressure improves conversion and protects catalyst, but compression represents significant energy cost. The relationship between hydrogen partial pressure and product quality is nonlinear, with diminishing returns at higher ratios. Models that learn unit-specific behavior can find the minimum hydrogen-to-oil ratio required to hit product specifications under current conditions, reducing consumption when feed quality permits while increasing it when heavier feeds arrive. This balance shifts continuously as feed and catalyst conditions change.

Product slate flexibility enables faster response to market conditions. Hydrocrackers can shift between maximizing middle distillates and maximizing naphtha depending on relative crack spreads. Traditional controls require manual intervention to adjust severity and fractionator setpoints. When optimization coordinates the entire unit simultaneously, it captures margin from market shifts that would otherwise require hours of operator attention. The ability to respond to crack spread changes within a shift rather than waiting for the next planning cycle represents a meaningful source of recovered value.

Building Confidence from Advisory to Closed Loop

Implementation follows a natural progression as trust develops between operators and the technology. The pathway matters because it determines whether the organization captures value or encounters resistance that stalls deployment.

Plants typically begin in advisory mode, where AI analyzes real-time data and provides recommendations while operators retain complete control. This stage delivers immediate value through enhanced visibility into process behavior and improved consistency across shifts. Operators compare recommendations against their own judgment, building the intuition needed to trust the system with more authority. When the AI suggests a temperature profile adjustment during a feed transition, operators can evaluate whether the recommendation aligns with their understanding of how the unit should respond. Advisory mode delivers returns on its own terms through better decision support and reduced variability across shifts and crews.

As confidence develops, plants progress to supervised operation. The AI executes setpoint changes within defined parameters while operators maintain oversight. Response time to feed changes improves because the system adjusts faster than manual intervention cycles allow. Process variability decreases as the technology responds to disturbances before they propagate through the unit. Operators focus on exception management rather than routine optimization, freeing attention for higher-value activities like troubleshooting equipment issues or coordinating with upstream and downstream units.

When Does Closed Loop Operations Come In

Closed loop operation represents the mature state where AI makes and executes optimization decisions continuously across multiple process variables. Operators retain full override capability and transition to strategic monitoring roles. The progression typically spans months to years, and the pace depends on organizational readiness, unit complexity, and the specific constraints each site faces. Some plants move quickly when operators see early results align with their expectations. Others proceed more gradually, building confidence through extended validation periods.

Some operations may remain in advisory mode indefinitely based on their specific constraints and objectives. Regulatory requirements, organizational culture, or labor agreements may favor human-in-the-loop operation. The technology delivers returns at each stage rather than requiring full automation before value materializes.

Capturing Hydrocracker Value with AI Optimization

For refinery operations leaders seeking to capture the margin improvements that traditional controls cannot deliver, Imubit’s Closed Loop AI Optimization solution offers a path forward. The technology learns from actual plant data and writes optimal setpoints to the distributed control system (DCS) in real time, capturing value from temperature profile optimization, quench distribution, and hydrogen management that static control strategies miss. Rather than requiring perfect data or immediate full automation, implementation begins with existing plant historian data and progresses at the pace each organization requires. Plants can begin in advisory mode, validating AI recommendations before progressing toward closed loop optimization as confidence builds.

Get a Plant Assessment to discover how AI optimization can recover hidden margin in hydrocracker operations.

Frequently Asked Questions

What data is needed to apply AI optimization to a hydrocracker?

Effective optimization requires historical process data from the plant historian covering reactor temperatures and pressures, hydrogen flows, product qualities, and operating conditions across multiple scenarios. Feed characterization data and lab results strengthen model accuracy. While richer datasets improve results, plants can begin with existing data and improve data quality iteratively as the system identifies gaps.

How does AI optimization handle the safety constraints that limit hydrocracker operations?

AI optimization operates within existing safety interlocks, alarm limits, and approved operating envelopes. The technology does not bypass or modify process safety systems. Instead, it finds optimal operating points within defined safe boundaries. Operators who validate recommendations through advisory mode build confidence that the system respects constraint boundaries while maximizing conversion.

How long before AI optimization delivers measurable results on a hydrocracker?

Plants typically observe measurable improvements within the first few months of advisory mode, including enhanced decision support and improved shift-to-shift consistency. The progression from advisory to closed loop spans months to years, with each phase delivering incremental value. Returns begin in advisory mode rather than requiring full automation for ROI.