Event-Based Switching Iterative Learning Model Predictive Control for Batch Processes With Randomly Varying Trial...

Area of Science:

• Control Engineering
• Artificial Intelligence

Background:

• Iterative learning model predictive control (ILMPC) excels at improving batch process tracking performance.
• Varying trial lengths in batch processes disrupt ILMPC by limiting prior information and control updates.

Purpose of the Study:

• To address the challenge of randomly varying trial lengths in ILMPC.
• To enhance the robustness and applicability of ILMPC in practical batch processes.

Main Methods:

• A prediction-based modification mechanism compensates for absent data in varying trial lengths.
• A 2-D neural network with adaptive parameters generates predictive compensation data.
• An event-based switching learning structure prioritizes recent trial data.

Main Results:

• The modified ILMPC guarantees convergence even with probabilistic trial length variations.
• The 2-D neural network provides accurate compensation for nonlinear processes.
• Simulations demonstrate the effectiveness on a numerical example and injection molding.

Conclusions:

• The proposed ILMPC modification and learning structure effectively manage varying trial lengths.
• The approach enhances control performance and learning stability in batch processes.
• Validated for complex nonlinear systems like injection molding.