论文部分内容阅读
Fisher判别分析(FDA)是一种有效的化工过程故障模式分类方法,但是其忽视了数据局部结构信息的挖掘。针对该问题,提出一种多块局部Fisher判别分析(MLFDA)方法,以更有效地识别化工过程故障。从变量和样本两个维度来分析数据的局部结构特性。针对变量维度的局部信息挖掘问题,设计了一种基于变量与数据集主元空间的相关度的变量分块方法,将全局过程变量划分为多个局部变量块。进一步考虑到样本维度的局部结构特性,应用基于局部权重因子的局部Fisher判别分析(LFDA)为每个局部变量块构建分类器。提出一种基于分类性能加权的多分类器集成方法,以融合不同分类器的决策结果。在Tennessee Eastman过程上的仿真结果说明,M LFDA方法具有比传统的FDA和LFDA方法更低的故障误分类率。
Fisher Discriminant Analysis (FDA) is an effective classification method for chemical process failure modes, but it ignores the mining of local structure information of data. Aiming at this problem, a multi-piece Local Fisher Discriminant Analysis (MLFDA) method is proposed to identify the chemical process faults more effectively. Analyze the local structural properties of data from the two dimensions of variables and samples. Aiming at the problem of local information mining with variable dimension, a variable block method based on the relativity between variables and data set’s principal component space is designed. The global process variables are divided into several local variable blocks. Further considering the local structural characteristics of the sample dimensions, a local Fisher discriminant analysis (LFDA) based on local weight factors is applied to construct a classifier for each local variable block. A multi-classifier integration method based on classification performance weighting is proposed to integrate the decision-making results of different classifiers. The simulation results on the Tennessee Eastman process show that the M LFDA method has a lower fault misclassification rate than the traditional FDA and LFDA methods.