maximisation of LCO from the FCC. A commonly employed strategy is to adjust operating conditions and reduce reactor temperature to preserve more LCO, coupled with a catalyst reformulation that offers better bottoms upgrading to avoid high slurry make at the low severity conditions. In this sce- nario, a refiner may also elect to use ZSM-5 additive to pre - serve overall volume swell and improve yield slate profitability. For refiners seeking to continually optimise FCC unit perfor - mance and maximise value, establishing a strong partnership with their catalyst supplier is essential. Various digitalisa- tion tools have been implemented to further enhance the operational agility of an FCC. These tools extend the unit’s inherent flexibility, enabling more responsive and precise decision-making. By supporting systematic evaluation and rapid implementation of optimised operating conditions and catalyst formulations, these solutions drive measurable per- formance gains, resulting in significant annual savings and improved margins. Beyond individual yield strategies, digital solutions play a pivotal role in enabling continuous optimisation. Seamless data exchange between FCC operators and catalyst suppliers allows for real-time monitoring and notification, empower - ing timely adjustments to catalyst formulations and process conditions. Simulation and optimisation tools further support this by evaluating alternative scenarios and catalyst options. These alternate scenarios enable refiners to proactively adjust operations to respond to market shifts such as declin- ing gasoline demand or rising petrochemical prices. These services also focus on capturing immediate value, helping to identify and act on optimisation opportunities quickly. In doing so, they support broader FCC adaptation strategies, ensuring refiners remain competitive throughout the run between turnarounds. OlefinsUltra is a mark of W. R. Grace & Co. A Edison Tan, Business Consulting Engineer, Imubit, edi- son.tan@imubit.com FCC-focused refiners are particularly vulnerable to rapid market changes due to the inherent flexibility and complexity of the FCC unit. Unlike hydrotreaters or distillation columns, FCC operations sit at the crossroads of feedstock variability, product value volatility, and unit non-linearity. As the eco- nomic heart of many complex refineries, the FCC is central to maximising refinery margins, especially in light ends and middle distillates. As such, they offer outsized margin oppor- tunities but also risk during volatile market cycles. In a four-year continuous run, refiners must expect shift - ing product spreads (for example, gasoline vs distillate), tightening environmental specifications, and feed avail - ability, all while contending with catalyst deactivation, coke constraints, and hardware reliability. Traditional planning strategies, driven by a monthly LP disconnect from live unit behaviour, will struggle to respond meaningfully. In an increasingly volatile oil environment, some refiners are already adapting by running their LP weekly, but limitations remain. Yield vectors are static and linear approximations; constraints are averaged or underrepresented. As a result, the gap between ‘LP intent’ and ‘unit execution’ remains wide,
From LP intent to execution reality: The disconnect
LP-generated plan
Actual unit behaviour
Inputs
Nonlinear outputs
Static yield vector
Constraints
Real constraints
Ignored constraints
Yield vectors
Margin loss
Planning- actual variance
Figure 1 Planning vs actual variances
creating persistent planning vs actual variances that erode margins and delay critical strategy shifts (see Figure 1 above). To stay agile, refiners must tightly couple market intel - ligence, planning tools, and unit execution. This means integrating daily price signals, feedstock availability, and downstream demand directly into decision-making at the pace of operations, not planning cycles. This is where refin - ers can lean on powerful advancements in AI technology to enable this transformation. One such AI technology is closed loop AIO, achieving min- ute-to-minute optimisation strategy execution of complex nonlinear processes through two key components: Dynamic Process Model and Reinforcement Learning (RL) Controller. The Dynamic Process Model is a high-fidelity, nonlinear simulation of the FCC built upon actual plant data and pro- cess knowledge. This ‘process engineering digital twin’ cap- tures the true behaviour of the unit, including non-linear yield shifts, riser severity impacts, volume gains and constraint interactions. The RL Controller agent is trained through hundreds of thousands of trial-and-error scenarios on the previ- ously mentioned simulator. In this offline environment, the RL learns to act as a controller and optimiser, adapting to changing prices, constraints, and feed compositions. These optimised targets are sent to key manipulated variables for FCC, such as heavy feed intake, riser outlet temperature, and MAB, flowing directly into the distributed control system (DCS) or APC. Unlike static optimisers, this dynamic closed- loop optimisation can update continuously within two to five minutes, enabling higher uptime and responsiveness to con- tinuously changing unit conditions. When applied across multiple units (such as the hydrotreater or coker), the approach also supports multi- unit optimisation. The RL layer built upon dynamic process models of multiple units accounts for interdependencies and constraints on adjacent units that conventional unit-level solutions often overlook. To equalise the dynamic closed-loop model with planning, engineers must derive accurate LP vectors, validate assump- tions, and provide feedback much faster, often within weeks instead of months. This is where accurate AI models can help. When deployed in closed loop, engineers can quickly extract recent and frequently executed gains as LP yield vec- tors for updates.
15
PTQ Q4 2025
www.digitalrefining.com
Powered by FlippingBook