独立元子空间算法及其在故障检测上的应用

Independent component subspace method and its application to fault detection

针对高维数据建模问题,提出一种独立元子空间算法(ICSM),作为一种新的集成学习算法,ICSM利用独立元在不同变量上的贡献度来选取子空间,符合了集成学习的要求,具备了明确的物理意义,有效地克服了随机子空间算法(RSM)的主要缺点。在此基础上,进一步将ICSM应用于工业过程监控,提出了一种新的ICSM-PCA故障检测算法。首先在各个子空间内分别建立相应的PCA监测模型,然后根据T^2和SPE统计量的值计算出集成时各自的权重,最后构造两个集成统计量对工业过程进行监测。通过在Tennessee Eastman(TE)模型上的仿真研究,说明提出的算法具有较好的建模效果和故障检测能力。

To handle the modeling problem for high-dimension data, the independent component subspace method （ICSM） was proposed. As a new ensemble learning method, ICSM could overcome the main drawback of the random subspace method. It constructed subspaces according to independent components （ICs） contributions on different process variables. As a result, the modeling requirement of the ensemble learning method was satisfied, and its physical meaning was also well presented. Moreover, a new fault detection method named ICSM-PCA was also developed. Firstly, PCA monitoring models were build on different subspaces, then the weighted value of each model was computed based on T2 and SPE statistics. Finally, two ensemble statistics could be built for monitoring industrial processes. A case study of the Tennessee-Eastman （TE） process illustrated that the proposed method showed good modeling performance and exhibited satisfactory fault detection ability.