3基于多向核主元分析青霉素生产过程在线监测.pdf

412 浙 江 大 学 学 报( ) Vol.41 o. 2 2007 2 Journal of Zhejiang University(Engineering Science) Feb. 2007 刘世成, 王海清, 李 平 ( , , 310027) : ( PCA),, .( MKPCA) . PenSim2.0 , 2 MKPCA ,T SPE . , PCA ,MKPCA ,, . MKPCA , . : ; ;; ; : T P206. 3 : A : 1008- 973X(2007) 02- 0202- 06 Online monitoring of Penicillin production process based on multiway kernel principal component analysis LIU Shicheng, WA G Haiqing, LI Ping (N ational L abor ato ry of I nd us tr ial Contr ol Technology , I nstitute of I ndus tr ial P roces s Contr ol , Zhej iang Univ ersity , H angz hou 310027, China Abstract: Traditional principal component analysis ( PCA) based methods have been proved to be unreliable when applied to nonlinear process monitoring and false alarms might occur. A method based on multiway kernel principal component analysis (MKPCA) was proposed to capture the nonlinear characteristics of normal batch processes. Based on PenSim2. 0, the threeway data of a fedbatch Penicillin fermentation process were spread along batch direction and normalized. T hen the MKPCA model of the monitored process was set up, and T 2 and SPE statistics and their control limits were calculated using the model. Simulation results showed that the proposed method had better monitoring performance compared with PCA, and that fewer false alarms and small fault detection delay were obtained when the method was used to detect the substrate feed rate and the agitator power faults. MKPCA can effectively extract the nonlin ear relationships among the process variables. Key words: principal component analysis (PCA) ; kernel principal component analysis (KPCA) ; multiway kernel principal component a