Advantages of Symmetry's Molecular Reduced Order Model (ROM)
Application Example Highlights
Molecular Reduced Order Model
Symmetry's Molecular Reduced Order Model (ROM) technology allows a molecular-based Linear Programming (LP) style solver to run inside any unit operation(s) within a rigorous flowsheet.
You can think of it as a unit operation(s) running off a neural network solution that is trained by Symmetry’s rigorous solution already regressed to plant data. Symmetry’s technology tracks molecular structures through a neural net environment.
The main advantages that molecular ROM offers are:
- A fast solution is built into the simulation flowsheet unit operations on a custom unit-by-unit selection basis.
- It will atomically mass balance across the full model.
- Specific units or sections of the simulation can be changed back into rigorous solution mode for detailed studies.
- Understands the blending of different types of feedstocks by internally training on PIONA basis disturbances.
Currently, molecular ROM solution speed is compromised as it is integrated into a rigorous flowsheet, thus making the overall model slower when compared to LP or machine learning technologies. Table 1 offers a quick comparison between Symmetry’s molecular ROM, LP and machine learning technologies.
Table 1. Technology comparison
While all solutions are expected to reconcile flowrates in order to keep a material balance and provide stable solutions compared to rigorous models, there are other important advantages in the molecular ROM technology.
Keeping the atomic mass balance allows tracking of minor species, such as Sulphur and Nitrogen throughout the flowsheet, that are important for catalytic poisoning calculations and Hydrogen balance. The atomic balance is also required in order to reconcile the energy balance in reactive systems.
Molecular ROM is expected to trend correctly with changes in the feed composition, even if the feed was not used in the training.
Molecular ROM is also capable of matching bulk properties due to its fundamental understanding of the underlying composition. On the other hand LP and Machine learning models track properties independently of composition so there are likely to be inconsistencies especially in reactive systems.
What can you do with Symmetry molecular ROM?
Symmetry’s molecular ROM offers different levels of deployment according to your specific needs. By switching reactors or individual unit operations into ROM the solution time is greatly reduced:
- Evaluating new data points, running case studies or regressions will be greatly improved since less time is required for the model to solve.
- Will allow new users to get familiarized with the model easier due to faster feedback from the model.
- Depending on the size of the flowsheet full optimization of a plant section would be possible at this stage (example below).
To illustrate some of the functionalities and potential of molecular ROM let’s look at a Naphtha Reformer (CCR) with a Hydrotreater (HT) example by changing the characteristics of the feed. The feed change is summarized in the following table.
Table 2. Changes in Naphtha Feed properties
Figure 1. Flowsheet for Naphtha Reformer Model
The solution time for the full flowsheet includes controllers to automatically adjust the Hydrotreater temperature for the Catalytic reformer feed sulfur and another controller to manipulate the Weight Average Inlet Temperature (WAIT) to set the product RON.
Symmetry’s Molecular ROM is trained based on the rigorous solution and the range of the training can be set by the user. For example, the following image shows the CCR trained variables and rages, notice that it includes training on the molecular structure of the fluid.
Figure 2. CCR training summary
The solution time for the model goes from 1065s (~18min), where the slowest unit operation overall is the CCR taking over 70% of the total solution time. Having both reactors in ROM leads to a solution time of 84s (1.4 min), with the CCR taking only 1% of the solution time.
The model is 92% faster by solving the reactors in ROM.
Table 3. Model solution time
With such a speed-up you might think that the accuracy of the model would be affected, but if we compare some of the main results in the flowsheet, we find a minimal loss of accuracy as shown in the following table.
Table 4. Results comparison Rigorous vs ROM
Looking into the Future
Molecular ROM realistically deals with the model speed and stability combined with accurate mass, atomic and energy balance to allow optimization of multiple continuous variables on individual plant sections such as the Naphtha reformer.
As molecular ROM is expanded into additional unit operations combined with other productivity tools in Symmetry the times' requirements for a realistic full refinery optimization becomes possible in the near future.
Please contact your local VMG office for more information.
Victor M. Rodriguez, Ph.D. & Glen Hay, M.Eng., P.Eng., VMG Calgary