基于多约束DTW的MPCA间歇过程监测方法

针对间歇过程固有的批次不等长问题,也为了克服传统解决批次间同步问题方法存在的数据浪费、扭曲原始过程变量的自相关及交叉相关关系的严重缺陷,提出基于多约束的动态时间规整(dynamic time warping,DTW)方法,按照轨迹中点与点的模式进行动态匹配解决的同步问题.同时,引入了全局路径限制和失真度阈值限制对DTW方法进行改进,解决了传统DTW方法长时间运行造成的故障监测严重滞后的问题,同时克服了其处理过程的复杂性与其离线性导致其实际应用的困难.用多向主元分析(multiway principal component analysis,MPCA)方法将多约束DTW处理过的数据进行建模.将该方法应用到青霉素发酵过程仿真实验中,结果表明:该方法能够快速准确地对不等长批次进行规整,与传统方法相比,故障的误报率、漏报率明显降低.

Batch Process Monitoring Method Based on Multi-way Principal Component Analysis and Limited-DTW

A limited-DTW method which makes dynamic match by tracking the points in the trajectory between batches was proposed to solve the inherent unequal length problem in the batch process. Meanwhile the serious shortcomings of traditional method in data wasting and the distorted auto-and-cross-correlation of original process variables were overcome. The global path constraint and the distortion threshold constraint were introduced to improve the dynamic time warping. The serious lag was overcome by the limited-DTW method and it was applied in practical production by reducing the calculation. Multiway principal component analysis was used to model for the processed data by limited-DTW in the passage. The method was applied to fed-batch penicillin fermentation process. Experiments show that this method can effectively warp unequal batch to equal length and reduce the leaking alarms and nuisance alarms,which also proves that the proposed method has more reliable monitoring performance than that of the traditional method.