PI-Based Set-Point Learning Control for Batch Processes With Unknown Dynamics and Nonrepetitive Uncertainties

• 成果类型: Article
• 署名作者: Hao, Shoulin; Liu, Tao; Rogers, Eric; Wang, Youqing; Paszke, Wojciech
• 署名单位: Dalian University of Technology; Dalian University of Technology; University of Southampton; Beijing University of Chemical Technology; University of Zielona Gora
• 刊物名称: IEEE TRANSACTIONS ON AUTOMATIC CONTROL
• ISSN/ISSBN: 0018-9286
• DOI: 10.1109/TAC.2024.3512196
• 发表日期: 2025
• 页码: 3649-3664
• 关键词: convergence; Batch production systems; PROCESS CONTROL; uncertainty; Feedback control; Tuning; nonlinear dynamical systems; PI control; optimization; iterative methods; Batch processes with unknown dynamics; Data-driven control (DDC); iterative extended state observer (IESO); proportional-integral (PI) controller tuning; robust convergence analysis; set-point learning control

摘要

For industrial batch processes with unknown dynamics subject to nonrepetitive initial conditions and disturbances, this article develops a novel adaptive data-driven set-point learning control (ADDSPLC) scheme based on only the measured process input and output data, which has two loops, one for the dynamics within a batch and the other for the batch-to-batch dynamics. In the former case, a model-free tuning strategy is first presented for determining the closed-loop proportional-integral controller parameters. For the latter case, a set-point learning control law with adaptive set-point learning gain and gradient estimation is developed for batch run optimization. The robust convergence of the output tracking error is rigorously analyzed together with the boundedness of adaptive learning gain and real-time updated set-point command. Moreover, another iterative extended-state-observer-based ADDSPLC scheme is developed with rigorous convergence and boundedness analysis to enhance the robust tracking performance against nonrepetitive uncertainties. Finally, two illustrative examples from the literature are used to demonstrate the effectiveness and superiority of the new schemes over the recently developed data-driven learning control designs.