Iterative Learning Model Predictive Control Based on Iterative Data-Driven Modeling.

Area of Science:

• Chemical Engineering
• Control Systems
• Artificial Intelligence

Background:

• Iterative learning model predictive control (ILMPC) excels in batch process tracking but requires accurate models.
• Complex nonlinear batch systems often lack precise first-principles models, hindering traditional ILMPC.
• Data-driven approaches offer a solution for modeling and control in such systems.

Purpose of the Study:

• To develop a data-driven ILMPC for nonlinear batch processes using neural networks.
• To improve tracking accuracy and computational efficiency in batch process control.
• To address the challenge of unavailable accurate models in complex systems.

Main Methods:

• Utilized a control-affine feedforward neural network (CAFNN) to identify nonlinear batch system dynamics from process data.
• Formulated the ILMPC within a tube framework to mitigate modeling errors and ensure sustained accuracy.
• Leveraged the control-affine structure for offline analytical computation of objective function gradients.

Main Results:

• The proposed data-driven ILMPC demonstrated effective high-precision tracking for batch processes.
• The tube framework successfully attenuated the influence of modeling errors.
• Offline gradient computation improved online computational efficiency and optimization feasibility.

Conclusions:

• The data-driven ILMPC using CAFNN provides an effective solution for high-precision tracking in nonlinear batch systems.
• The method enhances control performance by learning system dynamics from data.
• Theoretical analysis confirmed the robust stability and convergence of the proposed control system.