文献类型：期刊文献

英文题名：A computationally efficient policy optimization scheme in feedback iterative learning control for nonlinear batch process

作者：Gao, Kaihua[1];Lu, Jingyi[2];Zhou, Yuanqiang[3];Gao, Furong[1,4]

机构：[1]Hong Kong Univ Sci & Technol, Dept Chem & Biol Engn, Kowloon, Hong Kong, Peoples R China;[2]East China Univ Sci & Technol, MOE Key Lab Smart Mfg Energy Chem Proc, Shanghai 200237, Peoples R China;[3]Tongji Univ, Coll Elect & Informat Engn, Shanghai 201804, Peoples R China;[4]Guangzhou HKUST Fok Ying Tung Res Inst, Guangzhou 511458, Peoples R China

年份：2025

卷号：195

外文期刊名：COMPUTERS & CHEMICAL ENGINEERING

收录：;EI(收录号：20250517772786);WOS:【SCI-EXPANDED(收录号:WOS:001413308200001)】；

基金：This work was supported in part by National Natural Science Foundation of China under Grant NO. 62394343, in part by the Hong Kong Research Grant Council under Grant 16203322 and Grant N_HKUST628/22; in part by the National Science Foundation of China under Grant 62403359; in part by the Fundamental Research Funds for the Central Universities, China, under Grant 22120240010.

语种：英文

外文关键词：Batch process control; Iterative learning control; Policy optimization; Gaussian process

摘要：In this paper, we propose a computationally efficient feedback iterative learning control (ILC) scheme for nonlinear batch processes. We present a structured framework that delineates the feedback ILC as a composite of two integral components: a state feedback controller and a conventional ILC mechanism. Within this framework, we employ policy search techniques to optimize the feedback component. In parallel, we tackle the feedforward aspect by formulating a stochastic optimal ILC problem. These two components are offline iteratively updated, thereby ensuring convergence under ideal conditions. To account for missing process models in practical scenarios, we incorporate Gaussian process (GP) modeling into our framework. By leveraging the GP model, we extend our iterative optimization approach to a GP-based feedback ILC optimization algorithm that guarantees tractability. We use two numerical examples to demonstrate the merits of our framework, including its fast convergence and effective rejection of disturbances.