面向间歇过程的迭代学习预测控制理论和算法研究

As a perferred choice for manufacturing of low-volume and high-value products, batch processes play an increasingly important roles in modern industries,such as plastic processing, specialty chemicals and pharmaceuticals. Batch processes have the following unique features comparing to continuous processes: absence of steady state, strong nonlinearties, limited cycle time, repetitive nature and multi-rate sampling data. To deal with the forementioned difficulties, iterative learning model predictive control algorithms are investigated and developed for batch processes. This project focus on the following research topics:　1)nonlinear iterative learning model predictive control based on parameterized LTV model;2) adaptive iterative learning model predictive control based on 2D recursive identification; 3)multi-rate iterative learning model predictive control. Final, the effectiveness and robustness of the proposed methods will be verfied exxpermentally on industral injection molding machine.The aim of this project is to enrich the existing control theory of batch processes and improve the automation level of batch processes in china.

为了适应瞬息万变的市场需求，现代工业正逐渐倚重于生产小批量、多品种、高附加值产品的间歇过程，如塑料加工、精细化工、生物制药等众多领域。与连续过程不同，间歇过程存在非稳态操作、非线性时变、有限运行时间、运行重复性、多速率反馈等鲜明特点。综合模型预测控制和迭代学习控制的优缺点，本课题拟研究面向间歇过程的迭代学习预测控制理论和算法体系，重点研究基于参数化LTV模型的非线性迭代学习预测控制、基于2D递归辨识的自适应迭代学习预测控制、间歇过程的多速率迭代学习预测控制，并在注塑机工业平台上加以应用验证。通过本项目的研究，丰富现有间歇过程的控制理论和方法，提高我国间歇生产过程的自动化水平。

为了适应瞬息万变的市场需求，现代工业正逐渐倚重于生产小批量、多品种、高附加值产品的间歇过程，如塑料加工、精细化工、生物制药等众多领域。与连续过程不同，间歇过程存在非稳态操作、非线性时变、有限运行时间、运行重复性、多速率反馈等鲜明特点。综合模型预测控制和迭代学习控制的优缺点，本课题开展面向间歇过程的迭代学习预测控制理论和算法研究，并提出基于时间参数化LTV模型的迭代学习预测控制方法、面向间歇过程的高效2D预测函数控制算法、面向间歇过程的二维递归最小二乘辨识算法、基于MVU滤波器的无偏模型预测控制算、基于收敛深度控制的嵌入式预测控制算法等研究成果，并在注塑机工业平台上加以应用验证。通过本项目的研究，丰富了现有间歇过程的控制理论和方法，为实际应用的开展提供方法及技术支撑。