Novel Framework for Simulated Moving Bed Reactor Optimization Based on Deep Neural Network Models and Metaheuristic...

Area of Science:

• Chemical Engineering
• Process Optimization
• Artificial Intelligence in Chemical Processes

Background:

• Model-based optimization of simulated moving bed reactors (SMBRs) is computationally intensive.
• Surrogate models, particularly artificial neural networks (ANNs), are increasingly used for complex process modeling.
• ANNs have been applied to simulated moving bed (SMB) units but not extensively to reactive SMBs (SMBRs).

Purpose of the Study:

• To develop and evaluate deep recurrent neural networks (DRNNs) for optimizing simulated moving bed reactors (SMBRs).
• To establish a consistent method for assessing optimality using surrogate models in SMBR optimization.
• To characterize the feasible operation region of SMBRs using DRNNs.

Main Methods:

• Optimization of SMBRs using deep recurrent neural networks (DRNNs).
• Characterization of the feasible operation region by recycling data points from a metaheuristic technique for optimality assessment.
• Utilizing ANNs for modeling the SMBR unit.

Main Results:

• DRNNs demonstrate capacity for modeling and optimizing SMBRs.
• The proposed method successfully characterizes the feasible operation region for SMBRs.
• The DRNN-based optimization approach meets optimality criteria for complex SMBR processes.

Conclusions:

• Deep recurrent neural networks (DRNNs) provide an effective solution for computationally demanding SMBR optimization.
• The integration of DRNNs with metaheuristic techniques offers a robust method for optimality assessment and feasible region characterization.
• This study advances the application of ANNs in reactive separation processes, paving the way for more efficient chemical process design.