Every refinery runs a version of the same daily problem. The crude distillation unit (CDU) shifts cut points in response to a cargo change. Within hours, catalytic cracker feed quality drifts, hydrotreater severity needs adjusting, the hydrogen balance tightens, and the gasoline blend pool moves off target.
These interactions keep moving through the site. The margin they cost sits between units, between planning and execution, and between shifts. Refineries with similar equipment can perform differently, and most of that difference shows up in execution: one unit changes conditions, and the units downstream absorb the impact without a shared economic signal.
Where margin leaks between units
Margin leaks at the boundaries between units, where decisions made for local reasons cascade downstream with no economic signal traveling with them. A refinery is a tightly coupled system. The crude unit produces straight-run cuts that need secondary upgrading before they meet transportation-fuel specifications, and every conversion and treating unit downstream exists to close that gap.
How local decisions cascade
Suboptimization at any point degrades margin downstream. Two examples:
- When a vacuum distillation unit (VDU) sees sudden steam surges that disrupt vapor-liquid contact, lighter distillates slip into the bottoms residue. That is a direct yield loss, and it constrains every unit that receives vacuum gas oil (VGO) feed.
- When operators raise FCC riser outlet temperature to make more light olefins, dry gas make rises too. That can strain wet gas compressor capacity, and it shifts the octane contribution to the gasoline blend pool.
Operators know these interactions unit by unit. The harder part is that no single controller spans the chain from the CDU through the FCC to the blend header. Cut-point decisions made for CDU reasons propagate downstream without a coordinated signal to the units that absorb the consequences. Catalyst aging, exchanger fouling and feed quality changes accumulate the same way. Each is visible to the unit engineer responsible for it, but the effect on refinery economics is rarely visible in real time.
Why blending locks in the loss
Blending sits at the end of the chain as an irreversible integration point. Octane targets set conservatively above specification lock in giveaway that upstream optimization cannot recover once the material reaches the blend header.
Why planning models and APC leave gaps in execution
Planning models and APC work on different time horizons and different objectives from real-time, cross-unit economics. Planners run offline linear program (LP) models to build average operating plans over weeks to months, and schedulers break those plans down in spreadsheets. Real-time operations then respond to conditions that neither the plan nor the schedule anticipated.
How plans and control drift apart
LP solutions land at intersections of constraint boundaries. When operations shift to different product qualities or a different schedule, the shadow prices are no longer valid. Console operators end up making real-time economic decisions on stale signals, or on none.
Advanced process control (APC) keeps individual units within their target operating ranges and is widely used to improve unit performance. Conventional model predictive control (MPC) runs each unit's controller independently, targeting local specifications from estimates and delayed feedback from adjacent units.
Maintenance adds to the problem. Large-matrix MPC needs specialized APC engineers for ongoing model calibration, and many of those engineers have moved into other roles. Where site skills thin out, controllers can drift into a degraded or inactive state without anyone noticing the performance loss.
How shift variation widens the gap
The same equipment, run under similar feed and ambient conditions, can produce noticeably different yields from one shift to the next. The spread reflects different operator strategies, different comfort with pushing constraints, and different responses to the same alarm patterns. Narrowing it is an opportunity in day-to-day execution that needs no new equipment.
How cross-unit optimization closes the gaps
Cross-unit optimization decides what setpoints each unit should target and updates those decisions as plant conditions change. APC handles setpoint execution well. Day-to-day operations also need continuous economic optimization across units, sitting above APC.
Why linear models fall short on real plants
Conventional APC relies on linear, empirical model approximations that hold within narrow operating windows. Those approximations lose accuracy when process gains shift with crude slate changes or catalyst aging. Models trained on a refinery's own operating history can capture nonlinear relationships that linear MPC cannot represent.
A catalyst activity improvement in a treating unit, for example, can support a feed rate increase, a longer cycle or better product quality, but one at a time. A model with visibility across units can evaluate which use generates the most value under current refinery-wide conditions. That kind of hydrocracker yield optimization is hard to do with separate linear models for each piece of equipment.
How advisory mode builds operator trust
Cross-unit recommendations deliver value when console operators act on them, and trust usually builds when a deployment starts in advisory mode. The model recommends setpoint changes, and operators decide whether to act. Crews compare the recommendations against unit behavior they already know, and trust grows through repeated accuracy.
In well-run advisory deployments, operators start to challenge the model. They run a recommendation, compare the result with the move they would have made, and learn from the cases where the model caught a cross-unit trade-off they had not weighed. Shift supervisors accept or override moves and record their reasoning when they deviate. Over time that record feeds back into the model, and incoming shifts inherit a clearer picture of what was tried, what worked and why. Yield variation between shifts narrows as crews converge on a shared reference for plant behavior, through closer human-AI collaboration.
The same reference serves planning, operations and engineering, so console operators and planners stop working from different versions of the same plant. Once crews trust the moves, the plant can progress to closed loop, with the controller writing setpoints, inside limits set by plant engineers.
Working with existing control
Cross-unit optimization works as a layer that decides which setpoints to target, while the existing control system continues to handle execution within each unit. Plants improve coordination without changing their control system structure. The added value comes from linking real-time decisions to refinery-wide economics, and from keeping that link current as crude slate, catalyst activity and product specifications change.
What performance looks like when the gaps close
Refinery performance turns on five levers: yield, giveaway, specific energy consumption, throughput and reliability. The gap between average and best-quartile performance on those levers is recoverable margin, and it compounds when units coordinate.
Where returns show up early
Giveaway reduction is often the earliest return. Static buffers applied independently at each unit accumulate into conservative product targets. Optimization can tighten them by adjusting setpoints to the actual blend composition in real time, instead of holding fixed margins, without changing any physical assets.
Throughput and yield gains build on that. Closing the gaps often surfaces capacity that was always there but never used, because no one could see which constraint was binding under current conditions. Energy efficiency gains add to the total, particularly in furnace operations and steam balance.
Why operating discipline compounds
Cash operating costs vary across the refining peer group with operating discipline. The differences come from how consistently units coordinate, how quickly economics reach the control room, and how well optimization holds across shifts. Closing those gaps is mostly about the quality and timeliness of decisions made on equipment already in place.
The same day-to-day data also feeds longer-cycle decisions. A model that tracks unit-level degradation in real time gives planning teams visibility into trends that LP models update infrequently. That can change how turnaround scope is defined, how heavy a crude slate the refinery commits to, and how hard to push unit constraints between major maintenance windows.
How Imubit applies this
In refining, Imubit applies Map, Model and Control to the margin between units. Map traces each leak to the recurring decisions behind it, such as cut points, reactor severity and blend targets. Model uses Deep Learning Process Control® (DLPC®) to learn each unit's nonlinear dynamics from its own history, guided by first principles. Control takes those decisions to closed loop, writing setpoints to the existing APC and DCS minute by minute, inside limits set by plant engineers. In refining, deployments have delivered a $0.25/bbl margin improvement, and at one refinery's naphtha and jet complex, closed-loop control of three fractionators delivered a 2 to 3% increase in jet recovery on top of existing APC. Read more about refining.
