文献类型：期刊文献

英文题名：Independent exponential slow feature analysis for fine-scale monitoring multimode processes: Application to nonstationary crude distillation units

作者：Long, Jian[1];Jiang, Siyu[1];Zhai, Jiazi[2];Ma, Weiwei[2];Ma, Wei[3];Li, Zhi[1]

机构：[1]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[2]China Petr Pipeline Bur Co Ltd, Langfang 065000, Peoples R China;[3]SUPCON Technol Co Ltd, Hangzhou 310000, Peoples R China

年份：2025

卷号：316

外文期刊名：CHEMICAL ENGINEERING SCIENCE

收录：;EI(收录号：20252218533926);WOS:【SCI-EXPANDED(收录号:WOS:001502916700003)】；

基金：This work was supported by National Natural Science Foundation of China (62394343, 62373155) , the Natural Science Foundation of Shanghai under Grant 24ZR1415900, the Shanghai Pilot Program for Basic Research (22TQ1400100-16) and the State Key Laboratory of Industrial Control Technology, China (Grant No. ICT2024A26) , Fundamental Research Funds for the Central Universities (222202517006) .

语种：英文

外文关键词：Crude Distillation Unit; Working condition recognition; Process monitoring; DBSCAN Algorithm; Independent Exponential Slow Feature Analysis

摘要：Ensuring the stability and efficiency of industrial processes necessitates the timely identification and accurate monitoring of process states. Traditional methods, such as global statistical models and supervised clustering, struggle to handle non-stationary conditions and dynamic operating modes. To overcome these limitations, this study proposes a condition-driven multimode monitoring framework. XGBoost, combined with operational expertise, identifies condition-driven variables highly correlated with performance indicators. A clustering algorithm then segments operating states into distinct modes. Building on the exponential slow feature analysis (ESFA) method, givens rotation is incorporated to enhance feature independence, resulting in the improved independent exponential slow feature analysis (IESFA) algorithm. This enhanced model constructs static and dynamic monitoring indicators, with control limits calculated using kernel density estimation, improving monitoring accuracy. A case study of a crude distillation unit demonstrates the framework's effectiveness in managing multimode and non-stationary processes. Future work will focus on enhancing real-time adaptability and broadening industrial applications.