迭代学习模型预测控制研究现状与挑战  

Status and Challenges of Iterative Learning Model Predictive Control

作　　者：马乐乐 刘向杰[1] 高福荣[2] MA Le-Le;LIU Xiang-Jie;GAO Fu-Rong(School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China;Department of Chemical and Biomolecular Engineering,Hong Kong University of Science and Technology,Hong Kong 999077,China)

机构地区：[1]华北电力大学控制与计算机工程学院,中国北京102206 [2]香港科技大学化学与生物分子工程学系,中国香港999077

出　　处：《自动化学报》2022年第6期1385-1401,共17页Acta Automatica Sinica

基　　金：国家自然科学基金(62073136,U1709211);中国博士后科学基金(2021M701184);国家重点研发计划(2021YFE0190900)资助。

摘　　要：历经20多年的发展,迭代学习模型预测控制在理论和应用方面都取得了长足的进步.但由于批次工业过程复杂多样、结构各异、精细化程度较高,现有的迭代学习模型预测控制理论仍面临着巨大挑战.本文简要回顾了迭代学习模型预测控制理论的产生及发展,阐述了二维预测模型、控制律迭代优化及二维稳定性等基本理论问题;分析了现有方法在理论及应用方面的局限性,说明了迭代学习模型预测控制在迭代建模、高效优化、变工况适应等方面面临的难点问题,提出了可行的解决方案.简要综述了近年来迭代学习模型预测控制理论和应用层面的发展动态,指出了研究复杂非线性系统、快速系统、变工况系统对进一步完善其理论体系和拓宽其应用前景的意义,展望了成品质量控制和动态经济控制等重要的未来研究方向.After more than 20 years of development,iterative learning model predictive control(ILMPC)has made great progress in the aspects of theory and application.Since batch processes are featured by strong complexity,diversity,structural variety and high degree of refinement,the existing theories are still faced with great challenges.This paper reviews the generation and development of ILMPC,and expounds the basic theory of two-dimensional predictive model,iterative optimization of control law and two-dimensional stability;The limitations of existing methods are analysed from the perspectives of theory and application,and the issues on iterative modelling,efficient optimization and variable operation conditions are discussed in detail with available solutions provided.The new trends in the theory and application of ILMPC are comprehensively reviewed.The studies on complex nonlinear system,fast system and off-design system are clarified to make great significance for further perfecting ILMPC theory and broadening its application.The important future research directions consisting of end product quality control and dynamic economic control are outlined.

关 键 词：迭代学习模型预测控制 二维预测模型 控制律迭代优化 复杂非线性系统 快速系统 变工况 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]