间歇过程最优迭代学习控制的发展:从基于模型到数据驱动

本文综述了间歇过程的基于模型的和数据驱动的最优迭代学习控制方法.基于模型的最优迭代学习控制方法需要已知被控对象精确的线性模型，其研究较为成熟和完善，有着系统的设计方法和分析工具.数据驱动的最优迭代学习控制系统设计和分析的关键是非线性重复系统的迭代动态线性化.本文简要综述了基于模型的最优迭代学习控制的研究进展，详细回顾了数据驱动的迭代动态线性化方法，包括其详细的推导过程和突出的特点.回顾和讨论了广义的数据驱动最优迭代学习控制方法，包括完整轨迹跟踪的数据驱动最优迭代学习控制方法，提出和讨论了多中间点跟踪的数据驱动最优点到点迭代学习控制方法，和终端输出跟踪的数据驱动最优终端迭代学习控制方法.进一步，迭代学习控制研究中的关键问题，如随机迭代变化初始条件、迭代变化参考轨迹、输入输出约束、高阶学习控制律、计算复杂性等.本文突出强调了基于模型的和数据驱动的最优迭代学习控制方法各自的特点与区别联系，以方便读者理解.最后，本文提出数据驱动的迭代学习控制方法已成为越来越复杂间歇过程控制发展的未来方向，一些开放的具有挑战性的问题还有待于进一步研究.

Optimal Iterative Learning Control of Batch Processes: From Model-based to Data-driven

A brief overview on model-based optimal iterative learning control (ILC) and data-driven optimal ILC for batch processes is presented. Model-based optimal ILC relies on an exactly known linear model. There are many systematic methods and tools for the optimal ILC controller design and analysis. The foundational of design and analysis tool of data-driven optimal ILC methods for nonlinear repetitive processes is iterative dynamic linearization. This work briefly reviews the model-based optimal ILC with its latest development. The data-driven iterative dynamic linearization method is revisited in detail with its properties and distinct features. The general data-driven optimal iterative learning control, including data-driven optimal ILC for a complete trajectory tracking, data-driven optimal point-to-point ILC for multiple intermediate points tracking, and data-driven optimal terminal ILC for the terminal output tracking, is overviewed and discussed. The key issues in terms of research of optimal ILC, such as stochastic initial conditions, iteration-varying reference trajectory/points, input and output constraints, high-order learning laws, and computational complexity are also presented and discussed. Moreover, this paper highlights and compares the model-based optimal ILC and the generalized data-driven optimal ILC, and demonstrates their relation and difference to facilitate general understanding of these methods. Finally, it is shown that the data-driven ILC methods are receiving increasing interest owing to the increasing complexity of batch processes. Some corresponding challenging problems are presented as well.