摘要详情

ID / 提交时间

16 / 2025-03-19 15:58:19

标题

Parallel Principal Entropy Component Analysis for Nonlinear Process Monitoring

关键字

principal component analysis, kernel entropy component analysis, parallel structure, feature extraction, nonlinear process monitoring

主题及专题

T02 数据驱动的故障诊断

状态

全文待审

作者

李涛 / 北京化工大学

韩永明 / 北京化工大学

马波 / 北京化工大学

耿志强 / 北京化工大学

摘要

Nonlinear process data usually contains multiple feature attributes, and it is challenging to deeply extract feature information from the data to achieve better process monitoring. To deeply extract the feature information in process data, the method of parallel principal entropy component analysis (PPECA) is proposed for nonlinear process monitoring in this paper. Specifically, the principal component analysis (PCA) is applied to downscale the data and remove the correlation of the data. The principal component and residual scores are calculated to achieve the first level of data feature mining. The single-layer feature extraction makes it difficult to mine the depth of information in the data. Therefore, a two-layer feature parallel extraction model based on the kernel entropy component analysis (KECA) is constructed in the principal and residual spaces, which allows the second layer of features to be mined for capturing the entropy information of the data. Fusing feature information from different data spaces, two statistics are constructed for process monitoring. Applying the proposed method to the numerical example and the actual blast furnace ironmaking process, the simulation results further validate the effectiveness of the proposed method and show superior process monitoring performance.