文献类型：期刊文献

英文题名：A novel state-space model identification method from a behavioral system-theoretic perspective

作者：Liu, Qingyuan[1];Wang, Yibo[1];Liu, Tao[2];Li, Zhongmei[3];He, Xiao[1];Shang, Chao[1]

机构：[1]Tsinghua Univ, Dept Automat, Beijing 100084, Peoples R China;[2]Dalian Univ Technol, Key Lab Intelligent Control & Optimizat Ind Equipm, Minist Educ, Dalian 116024, Peoples R China;[3]East China Univ Sci & Technol, State Key Lab Ind Control Technol, Minist Educ, Shanghai 200237, Peoples R China

年份：2026

卷号：163

外文期刊名：JOURNAL OF PROCESS CONTROL

收录：;EI(收录号：20261720595346);WOS:【SCI-EXPANDED(收录号:WOS:001756899500001)】；

基金：** This work is supported by National Natural Science Foundation of China (Nos. 62373211 and 62327807) , and the Open Research Project of the State Key Laboratory of Industrial Control Technology, China (Grant No. ICT2025B10) .

语种：英文

外文关键词：Linear system identification; State-space model; Subspace identification; Behavioral systems theory; Rank-constrained programming; Closed-loop identification

摘要：As the mainstream methodology for identifying state-space models, subspace identification relies on the orthogonality assumption between data spaces, and could thus lead to unsatisfactory identification accuracy in finite-sample regime. To overcome this limitation, a novel state-space model identification method is proposed by leveraging the capability of behavioral systems theory in characterizing system dynamics with finite-length data trajectory. In virtue of the innovation-based data-driven output predictor (DDOP), a recent advance from this theoretical framework, the state-space model identification is converted into an innovation estimation problem followed by a model reduction step. To achieve better identification accuracy, an improved innovation estimation strategy incorporating low-rank prior is further proposed, formulated as a rank-constrained programming (RCP) problem and solved via the alternating direction method of multipliers (ADMM). Numerical and industrial dataset experiments demonstrate the superior modeling accuracy of the proposed method over existing subspace identification methods in both open-loop and closed-loop cases, with out-of-sample prediction error reduced by more than 23% on industrial benchmark datasets.