基于局部特征的多模态过程监控方法

Multimode Process Monitoring Based on Local Feature

多模态过程中各个模态均有不同的特征,因此模态数据的局部特征比全局特征更能有效、合理地表征实际化工过程。为利用多模态数据的局部特征,提出了基于数据局部特征的多模型方法(LFMM)用于多模态过程的监控。首先,离线阶段考虑到数据间的时序信息以及数据特征,利用不同时间窗内数据的变异系数(CV)完成多模态数据集的聚类;然后,考虑到不同模态的数据在空间分布上具有不同的疏密性特征,建模阶段利用局部离群因子(LOF)算法计算数据在其模态数据集中的局部密度,监控时将在线数据的局部密度作为统计特征,并构造全局概率指标用于多模态过程监控;最后,通过田纳西伊斯曼(TE)过程验证了本文方法的有效性。

Every mode has different features in a multimode process, so the local features of modal data can be more effectively than global features for the reasonable characterization of chemical process. In order to use the local characteristics of multimodal data,this paper proposes a local feature based multiple model method , called, Local Feature-based Multiple M o d e l （LFMM ）, for process monitoring. Firstly, the sequential information between data and the modal data features is utilized in the offline phase and the coefficient of variance of data in different time w i n d o w s is applied for the clustering of the training data of multimode process. In the latter model p h a s e , L O F algorithm is utilized to compute the local data density in their mode data set. In the online phase,by taking the local data density as statistic character,a n e w global probability index is established as a monitoring statistic for multimode process monitoring. Finally, T E process is adopted to verify the effectiveness of the proposed m e t h o d .