基于动态结构保持主元分析的故障检测方法

Fault detection method based on dynamic structure preservation principal component analysis

关键词:  动态结构保持主元分析  流形学习  相关性分析  特征提取  故障检测

Key words:  dynamic structure preservation principal component analysis  manifold learning  correlation analysis  feature extraction  fault detection

摘要:

为充分利用表征过程运行工况的数据特征信息,提高化工过程的故障检测性能,提出一种基于动态结构保持主元分析(DSPPCA)的过程故障检测方法。首先对原始数据采用变量相关性分析建立自回归模型,构建包含动态特征的数据集,进一步综合考虑主元分析法(PCA)和局部线性嵌入(LLE)流形学习算法中数据点之间的近邻关系,融合得出新的目标函数,同时,运用局部线性回归的方法获得高维样本的嵌入映射,特征提取后在特征空间和残差空间分别构造监控统计量进行故障检测。Swiss-roll数据集的降维结果及TE过程的仿真研究结果表明,DSPPCA算法可以取得较好的特征提取效果,具有较高的故障检测性能。

Abstract:

In order to make full use of the feature information of data in the chemical process, a fault detection method based on dynamic structure preservation principal component analysis was proposed to improve the performance and efficiency for fault detection. It firstly established auto-regression model through correlation analysis so that the dynamic feature sets could be obtained to characterize the original data. Furtherly, principal component analysis and locally linear embedding were fused together to obtain a new objective function. Besides,locally linear embedding algorithm could preserve the neighbor relationship between data collected. At the same time, local linear regression was used to find the projection that best approximated the mapping from high-dimensional samples to the embedding for on-line application furtherly. Statistics were constructed in the two spaces for process monitoring after feature extraction respectively. Simulation results of Tennessee Eastman process and Swiss-roll data show that DSPPCA-based method is more effective for feature extraction and process monitoring.