基于改进的MPCA与DTW方法及其在批过程故障诊断中的应用  

Batch process monitoring and fault detection based on an improved MPCA and DTW

作　　者：肖应旺[1] 徐保国[1] 

机构地区：[1]江南大学通信与控制工程学院,江苏无锡214036

出　　处：《计算机与应用化学》2005年第5期344-348,共5页Computers and Applied Chemistry

基　　金：国家科技攻关计划"先进控制与优化软件及综合自动化软件平台产业化关键技术"资助(2001BA204B01-03)

摘　　要：针对传统的多向主元分析(Multiway Principal component Analysis,MPCA)常会导致误诊断,且对批生产过程难以保证在线状态监测和故障诊断的实时性,提出了一种改进的MPCA与动态时间错位(Dynamic Time Warping,DTW)方法,该方法采用多模型非线性结构代替传统的MPCA单模型线性化结构,并利用对称式DTW算法解决了多元轨迹同步化的问题。将该方法应用到青霉素发酵批过程的在线故障监测中,结果表明它克服了MPCA不能处理非线性过程和实时性问题,并避免了MPCA 在线应用时预报未来测量值带来的误差,提高了批过程性能监测和故障诊断的准确性。An improved multiway principal component analysis (MPCA) and dynamic time warping (DTW) for on-line batch process monitoring and fault detection was proposed, using multi-model instead of single model and symmetrical DTW algorithm to synchronizing batches. The improved method was applied to monitoring fed-batch penicillin production, and the results show that the approach can avoid the many false alarms resulted from predicting the future values of the test batch. So, the proposed method realizes the online fault monitoring of batch and improves the accuracy of MPCA process monitoring.

关 键 词：多向主元分析 批过程 多模型 在线监测 动态时间错位 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]