基于ICA和LOF的故障检测

Fault Detection Based on LOF of Augmented Matrix in Independent Component Space

针对实际工业过程中的高斯与非高斯变量同时存在的问题,提出一种基于独立元分析(independent component analysis,ICA)和局部离群因子(local outlier factor,LOF)的故障检测方案。利用ICA算法提取数据独立元,加入独立元的时滞输入特性和时差输入特性构建成增广矩阵。运用LOF算法剔除训练数据增广矩阵中的离群点,排除离群点对建模的影响。用剩余数据建立LOF模型,并计算检测指标LOF的值,通过核密度估计计算控制限。将检测指标与控制限做对比,确定检测数据是否正常。将该方案用于田纳西伊斯曼过程,并分别与传统的ICA和LOF方法比较,仿真结果说明该方法通过构建独立元空间增广矩阵和剔除离群点,有效地提高了LOF的故障检测率,同时也降低了误报率。

Considering the concurrence issue of Gaussian and non-Gaussian variables in the practual industrial process,a fault detection scheme based on ICA and LOF was proposed.Independent component analysis(ICA)algorithm was used to extract the independent components of data.The time delay and time difference input characteristics of independent components were added to construct the augmented matrix.The local outlier factor(LOF)algorithm was used to eliminate the outliers in augmented matrix of the training data,and eliminate the influence of outliers on the modeling.The remaining data was used to build the LOF model,and the values of the detection index LOF were calculated.The control limit of LOF was calculated by kernel density estimation.The detection index was compared with the control limit to determine whether the detection data was normal.This scheme was used in the Tennessee-Eastman process and compared with the traditional ICA and LOF methods respectively.The simulation results show that this method effectively improves the fault detection rate of LOF and reduces the false alarm rate by constructing augmented matrix in independent component space and eliminating outliers.