中文摘要：

针对微生物制药的间歇生产过程中缓变故障难于监测的问题,提出了多向核熵成分分析（multi-way kernel entropy component analysis,MKECA）过程监测的新方法,克服了传统多向核主成分分析（multi-way kernel principal component analysis,MKPCA）方法在监控缓变故障时漏报率高的缺陷.该方法首先将3维历史数据按照本文所提的3步法进行预处理,然后通过核映射将数据从低维空间映射到高维特征空间,解决数据的非线性特性,并在高维特征空间依据核熵的大小对数据进行降维,使降维后的数据分布与原点成一定的角度,能够逼近原始间歇过程的数据分布.通过数值实例和实际工厂数据对方法进行验证.结果表明,MKECA方法具有更可靠的监控性能,能及时、准确地监测出故障,具有广阔的应用前景。

英文摘要：

It is difficult to monitor the slowly changing faults in a pharmaceutical microbial batch process. We pro- pose a process monitoring method based on multi-way kernel entropy component analysis （MKECA）. When the traditional multi-way kernel principal component analysis method is used to monitor slowly changing faults, the rate of missing reports is high. Our method overcomes this problem. The method preprocesses three-dimensional historical data by using the proposed three-step method. Then the preprocessed data are mapped from a low-dimensional space to a high-dimensional space to solve the problem of nonlinear character- istics of the data. At the same time, in the high-dimensional feature space, the dimension will be reduced according to the size of data kernel entropy. The after-reduction data can thus form a certain angle with the orig- inal point, and the distribution can be close to the original data distribution of the batch process. The experiment results from numerical examples and an actual factory illustrate that the MKECA method has a more reliable monitoring performance, and it can monitor faults in a timely and accurate manner. Therefore, the proposed method has prospects for broad application.