基于多阶段多向核熵成分分析的间歇过程故障检测方法

针对间歇过程的非线性、多阶段特性，提出一种基于多阶段多向核熵成分分析（multistage-MKECA，MsMKECA）的故障检测方法。针对间歇过程的多阶段特性，建立一种时序核熵主元关联度的矩阵相似性阶段划分方法，实现对间歇生产过程的多阶段划分；针对传统批次展开方式在线监控需要预估批次未来值的缺陷，进一步引入一种批次-变量三维数据展开方式建立每个阶段的MKECA非线性统计模型，实现对间歇过程的分阶段监控。最后对盘尼西林发酵过程开展仿真研究，结果表明所提方法能够比传统MKECA方法更为快速地进行故障检测。

Fault detection in batch process by multistage multiway kernel entropy component analysis

A fault detection method, i.e., multistage multiway kernel entropy component analysis (MsMKECA) was proposed on the basis of nonlinearity and multistage characteristics of batch process. First, in order to divide a batch process into multiple stages, a matrix similarity stage division method was constructed from correlation matrixes of the time-series kernel entropy components. Then, a batch-variable 3-D unfolding technique was introduced to build MKECA model in each stage and to monitor operations in each stage of the batch process, which overcame on-line monitoring impediments of requiring estimation on future values by conventional batch-wise unfolding technique. Simulation study on penicillin fermentation process showed that the proposed method can offer much faster fault detection than traditional MKECA.