The Shell Oil Deer Park refinery is in the middle of its digitalization and end-to-end optimization journey. One recent project focused on our catalytic reformer unit, which converts low-octane feed to a high-octane product called reformate and light hydrocarbons ranging from C1 to C5+.
In 2019, the site installed an online Raman analyzer on the reformer unit to reduce variability of reformate product octane. At the same time, we commissioned the Platform for Advanced Control and Estimation (PACE) — a next-generation advanced process control (APC) technology jointly developed by Shell and Yokogawa — to optimize the reformer product octane.
Before the installation of the online Raman analyzer and PACE controller, the unit was regulated manually, which required lots of moves from the board operators to keep product on-specification and within the reactors and heater constraints. Monitoring product quality involved sending a sample to the laboratory six hours after a reactor switch (a move necessary to allow catalyst regeneration) and then waiting three to six hours for the laboratory results. This meant the unit didn’t reach the target octane until nine to 12 hours after the reactor switch, resulting in a 0.7 octane giveaway for a given target.
The new PACE application, in conjunction with the Raman analyzer, has compressed the decision cycle to two hours from nine, and reduced product variability by 71%, cutting octane giveaway to 0.2 from 0.7.
Modeling And Control Design
The Deer Park refinery has used APC technology since the 1980s. When the Raman analyzer was installed, we decided to replace the legacy APC with PACE because of its unique features and added robustness. The reformer unit process is highly interactive and complex. PACE’s modeling flexibility and customizable event logic were a perfect fit for this unit.
The reformer unit’s four reactors and their associated heaters were modeled in PACE as one system using a concept called intermediate variables. The idea behind this concept is to create fast and accurate models that represent the process, i.e., a gray box model.
For instance, in modeling a heater and associated reactor, three independent inputs (heater-outlet-temperature controller setpoint, feed rate and feed temperature) were related to an intermediate variable (heater-outlet-temperature controller process variable).
The model accurately captures firing rate and feed impact on the reactor inlet temperature. The reformer reactor inlet temperature determines octane target and final product composition. The next step is to take the outlet temperature and scale it down (one-quarter each to the other three heaters). The last step is to capture and model all four reactors as a single system. This involves calculating another intermediate variable, the average temperature of all four reactors, which then is used to get RON octane.
Increasing the average temperature of the reactors by 1°F will result in a 0.3 octane rise. Board operators can easily understand this when they’re analyzing PACE performance and predictions.
Furthermore, if the online Raman analyzer is in maintenance mode, operators now can use the calculated reactors’ average temperature as a controlled variable by simply changing it from high and low constraint limits to a single setpoint. Previously, the unit was manually regulated by using average temperature and laboratory samples. The intermediate variable modeling is well suited for this application because fast and accurate models made the controller more robust with minimal deviation from the octane target.
Figure 1. Each of the four heaters typically faces these constraints.
The primary objective of the PACE controller is to keep octane to a given setpoint while balancing the heater and reactor temperature profile.
We set up heater deviation temperature modeling so that operations can adjust the heater and reactor loading profile. Typically, the first reactor runs at high temperature with the other reactors not quite as hot.
Because PACE enables users to customize the operator human-machine interface (HMI), we created a customized variable set so operators can easily access common and key variables that require frequent adjustments. This allows the operators to display most-used variables without switching back and forth between four reactor HMIs.
The operators decide which heater will be connected to the lead reactor by setting the desired heater deviation setpoint.
Various heater constraints, such as minimum oxygen and maximum burner pressure limitations, determine whether or not the controller can deliver the targeted octane as well as the desired heater and reactor temperature profile. Figure 1 shows typical constraints for one of the four heaters. If a constraint limits a heater, PACE will shift firing load to the heaters that aren’t limited. When that happens, PACE no longer can maintain the heater profile temperature setpoint. However, PACE will continue to keep octane at its setpoint if at least one heater isn’t constrained.
As already noted, reactor switching is necessary to allow catalyst regeneration. Operations requested automatic shutoff of PACE before a reactor switch is initiated. Fortunately, a very useful feature in PACE called event logic makes setting up event logics based on operator-enabled rules easy. The controller uses two event logic rules: the primary rule is to shut off when the opening of the reactor bypass valve exceeds 5% — the valve is at 0% normally or 100% before reactor switching takes place; and the secondary rule, which applies when the reactor bypass valve is in maintenance mode, activates shut off when hydrogen injection pressure and flow controllers are switched from manual to automatic mode.
After completion of a reactor switch, the operators put the controller back into active mode when octane lines out and average temperature stabilizes.
Using the online Raman analyzer together with PACE has enabled the site to minimize product octane giveaway by 71%. Figure 2 shows octane variability from a given target remains within ±0.2. Additional benefits include maximizing feed to the unit while keeping heaters at their design point and, after a reactor switch completes, reaching octane target within two hours instead of the previous nine to 12 hours.
Figure 2. Online analyzer and Platform for Advanced Control and Estimation technology have provided several significant benefits.
We have commissioned the Raman analyzer connection to the reformer unit stabilizer. The purpose of this controller is to separate reformate product and lighter components. We are using PACE to regulate stabilizer bottom reformate C5s and overhead benzene. The new PACE controller and octane inferential both use the reformer-feedstream Raman analyzer components as feedforwards.
Going forward, the entire site will go through its planned journey with the goal of optimizing the plant.
MOHAMED ABOKOR deploys and maintains PACE applications at Shell Oil refinery in Deer Park, Texas. Email him at [email protected].