Q‑Learning-Based Multivariate Nonlinear Model Predictive Controller: Experimental Validation on Batch Reactor for...

Area of Science:

• Chemical Engineering
• Control Systems
• Artificial Intelligence

Background:

• Batch reactors require precise temperature control for optimal performance.
• Traditional control methods often rely on accurate system models, which can be challenging for nonlinear processes.

Purpose of the Study:

• To introduce a novel Q-learning-based nonlinear model predictive control (QL-NMPC) framework.
• To enable model-free temperature control in batch reactors using reinforcement learning.

Main Methods:

• A reinforcement learning agent was trained in simulation to learn optimal control policies.
• The Q-learning algorithm, utilizing value iteration, was employed for model-free policy optimization.
• The learned control policy was implemented in real-time on a physical reactor using the NVIDIA Jetson Orin platform.

Main Results:

• The QL-NMPC framework demonstrated effective temperature tracking in the batch reactor.
• The model-free approach successfully optimized control strategies without explicit policy evaluation.
• Real-time implementation on the NVIDIA Jetson Orin platform validated the framework's practical applicability.

Conclusions:

• Reinforcement learning offers a powerful approach for controlling nonlinear batch processes.
• The QL-NMPC framework provides an effective, model-free solution for temperature control in batch reactors.
• This study highlights the potential of AI in advancing process control without system identification.