基于分步时空JITL-MKPLS的间歇过程故障监测  

Fault monitoring of batch processes based on substep time-space JITL-MKPLS

作　　者：高学金 孟令军[1,2,3,4] 王豪 高慧慧[1,2,3,4] GAO Xue-jin;MENG Ling-jun;WANG Hao;GAO Hui-hui(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China;Engineering Research Center of Digital Community,Ministry of Education,Beijing 100124,China;Beijing Laboratory for Urban Mass Transit,Beijing 100124,China;Beijing Laboratory of Computational Intelligence and Intelligent System,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部,北京100124 [2]数字社区教育部工程研究中心,北京100124 [3]城市轨道交通北京实验室,北京100124 [4]计算智能与智能系统北京市重点实验室,北京100124

出　　处：《高校化学工程学报》2021年第1期127-139,共13页Journal of Chemical Engineering of Chinese Universities

基　　金：国家自然科学基金(61803005,61640312,61763037);北京市自然科学基金(4192011,4172007);山东省重点研发计划(2018CXGC0608);北京市教育委员会资助。

摘　　要：针对多阶段时变的间歇过程难以用全局模型准确描述生产过程的动态变化及传统局部建模每个工作点都需要重新筛选样本建模导致计算量较大的问题,提出一种分步时空即时学习的局部建模策略。采用仿射传播(AP)聚类的方式对历史数据样本集中的数据进行初步分类,在当前输入样本数据到达后,确定当前样本数据所属的类别,在此类别所限定的子数据样本集中使用时间和空间相结合的即时学习策略确定出局部相似样本,建立多向核偏最小二乘监测模型。将该算法在青霉素发酵仿真数据和大肠杆菌发酵过程生产数据上进行验证,结果表明,所提方法不仅减少了不必要的计算量,还能够更加精准即时地进行故障监测。It is difficult to accurately describe the dynamic characteristics of multi-stage time-varying batch processes with global models,while every working point of traditional local models needs to re-select samples that results in high requirement of computation.A local modeling strategy based on substep time-space just-in-time learning(JITL)was proposed.The data in the sample set of historical data were preliminarily classified by an affinity propagation(AP)clustering method.When the current input sample data arrived,the category of these data was determined.The JITL strategy combining time and space was used to determine local similar samples in the subsample of the category.A multi kernel partial least squares(MKPLS)monitoring model was established using local similar samples.The effectiveness of the proposed method was validated via simulation of penicillin fermentation and E.coli production of interleukin-2.The results demonstrate that the proposed method reduces unnecessary calculation and can detect faults more accurately and instantaneously.

关 键 词：时空即时学习 分步 多向核偏最小二乘 局部建模 故障监测 

分 类 号：TH165.3[机械工程—机械制造及自动化]