Reinforcement Learning-Based Nonlinear Model Predictive Controller for a Jacketed Reactor: A Machine Learning Concept...

Area of Science:

• Chemical Engineering
• Control Systems
• Artificial Intelligence

Background:

• Batch reactors (BRs) are vital in pharmaceuticals and specialty chemicals, but managing exothermic reactions and preventing thermal runaway remains challenging.
• Existing control methods struggle with the complex dynamics and diverse operational conditions inherent in batch processes.

Purpose of the Study:

• To develop and experimentally validate an integrated Machine Learning and Nonlinear Model Predictive Control (NMPC) framework for accurate temperature tracking in batch reactors.
• To enhance process safety, efficiency, and reduce energy consumption through intelligent control.

Main Methods:

• Utilized a Recurrent Neural Network (RNN) for open-loop modeling of lab-scale batch reactor data.
• Implemented an actor-critic reinforcement learning (A2CRL) methodology for dynamic weight updates within the NMPC framework.
• Optimized coolant flow rate dynamically to ensure precise temperature regulation and stability.

Main Results:

• The A2CRL-enhanced NMPC framework demonstrated improved controller performance compared to existing deep learning NMPC methods.
• Achieved precise temperature regulation, enhanced process efficiency, and reduced energy consumption.
• Validated the framework's potential for industrial-scale batch reactor systems, improving operational safety.

Conclusions:

• The proposed A2CRL-NMPC approach offers a robust solution for managing complex batch reactor dynamics.
• This methodology balances prediction accuracy with real-time computational efficiency for industrial applications.
• The successful experimental validation highlights its potential to significantly improve safety and reduce energy usage in chemical processing.