基于改进核主元和支持向量数据描述故障检测

Fault Detection Based on Improved Kernel Principal Component and Support Vector Data Description

提出基于改进核主元和支持向量数据描述(SVDD)故障检测方法,适合于复杂工业过程具有非线性和非高斯性的情况。首先,通过对核主元(KPCA)特征空间样本进行重构误差,在样本集上自动识别异常值,减少对KPCA算法的影响并增强非线性核映射。然后,利用支持向量数据描述算法处理数据非高斯信号,据此构建统计量对工业过程进行检测。最后,将所提出的改进核主元和支持向量数据描述方法应用于田纳西-伊斯曼(TE,Tennessee Eastman)过程的仿真实验,结果说明提出方法的有效性。

The method of fault detection based on improved kernel principal component and support vector data description（SVDD） is proposed, which is suitable for complex industrial processes with nonlinear and non Gauss. First of all, based on the reconstruction error of kernel principal component analysis （KPCA） feature space sample, the outliers in the sample set are automaticlly identified to reduce the impact on the KPCA algo- rithm and enhance the nonlinear kernel mapping. Then, the support vector data description algorithm is used to process non-Gauss signal, whereby to construct statistic to detect the industrial processes. Finally, the proposed improved kernel principal component and support vector data description method is applied to the Tennessee Eastman （TE） simulation process. The results show the effectiveness of the proposed method.