基于主元分析的支持向量数据描述的故障检测

Fault Detection of Support Vector Data Description Based on Principal Component Analysis

针对连续过程中故障检测数据量偏大、传统的支持向量数据描述(support vector data description,SVDD)检测时间长、检测效率低等不足,提出了一种主元分析(principal component analysis,PCA)与支持向量数据描述结合的故障检测方法.首先将原始数据用主元分析进行降维,获得维数简约的主元空间及残差空间;然后对主元空间和残差空间的得分矩阵先后运用支持向量数据描述方法建模,获得阙值;最后对新测试样本进行故障检测.以数值例子和TE过程数据进行仿真研究,实验结果表明该方法具有节约时间、降低漏检率的优点.

According to the data of continuous process fault detection in large amount,traditional support vector data description long detection time,low detection efficiency,a fault detection method based on principal component analysis and support vector data description is proposed.First,the original data is reduced by principal component analysis to obtain simple dimension principal component space and residual space.Then,the score matrix of principal component space and residual space is modeled using support vector data description method to get the threshold.Finally,fault detection of the test samples was conducted.By numerical example and TE process data,the experimental results show that proposed method possesses the advantages of saving time and the reduction of the residual rate.