Deep Learning-Based Model Predictive Control for Continuous Stirred-Tank Reactor System.

Area of Science:

• Chemical Engineering
• Control Systems
• Artificial Intelligence

Background:

• Continuous stirred-tank reactor (CSTR) systems are crucial in wastewater treatment but challenging to control due to identification difficulties.
• Accurate system modeling is essential for effective process control and optimization.

Purpose of the Study:

• To propose a deep learning-based model predictive control (DeepMPC) for accurate modeling and control of CSTR systems.
• To address the limitations of traditional system identification methods in complex industrial processes.

Main Methods:

• Developed a DeepMPC framework integrating a growing deep belief network (GDBN) for system identification and quadratic optimization for control.
• Utilized transfer learning within GDBN for automatic network sizing and improved predictive modeling.
• Analyzed the convergence and stability of the proposed DeepMPC algorithm.

Main Results:

• The GDBN model accurately approximated CSTR dynamics with bounded error, demonstrating high performance in system identification.
• DeepMPC exhibited superior performance in modeling accuracy, setpoint tracking, and disturbance rejection compared to state-of-the-art methods.
• Validated the effectiveness of DeepMPC on a second-order CSTR system.

Conclusions:

• DeepMPC offers a robust and effective solution for modeling and controlling CSTR systems, overcoming traditional identification challenges.
• The proposed deep learning approach significantly improves control performance in wastewater treatment applications.
• This work advances the application of AI in industrial process control and optimization.