基于仿射传播聚类子集主元分析的间歇过程监测方法

摘要

针对间歇过程固有的多阶段特性,也为了克服传统阶段划分方法严格按照物理时刻顺序将采样点硬性分割而不能使其寻找数据特征最为相近的聚类中心的严重缺陷,提出基于仿射传播聚类(AP)的子集多向主元分析(subset-MPCA)监测新方法:采用全新的乱序聚类思想,将时间片矩阵打乱用AP进行无约束乱序聚类,使样本突破时间顺序的约束自由找寻与其特征最为相近的聚类中心,获得聚类子集,建立精确的子集MPCA监控模型.在线监控时,引入信息度传递实现实时采样点的阶段归属判断,解决阶段不等长批次的最佳模型选择问题.对青霉素仿真数据的实验表明,该方法较传统方法可有效降低故障的漏报和误报,有着更加可靠的监控性能.%For the multiphase property inherent in the batch process, and in order to overcome the serious deficiency that the traditional stage partition method divided the sampling points into several categories strictly according to the sampling time sequence and cannot make it look for the clustering center with the most similar data characteristics, a novel subset-MPCA method based on affinity propagation (AP) clustering is proposed. Using a new idea of random order clustering, this method disrupts the order of the time slice and AP is used to cluster with the random order. Thus, each data point can break the restriction of the time sequence and find the clustering center which has the most similar data characteristics with it, obtaining clustering subset and establishing the precise model. For online monitoring, information transmission is introduced to determine the stage attribution of real time sampling points to solve the problem of optimal model selection for unequal length batch. Experiments on penicillin simulation data show that this method can effectively reduce the leaking alarms and nuisance alarms than the traditional method, having more reliable monitoring performance.