基于全局和局部信息融合的SVM多模态过程故障检测

SVM for Fault Detection of Multi-Modal Process Based on Global and Local Information Fusion

为了有效地提高支持向量机(SVM)对多模态过程的故障检测性能,提出一种基于全局和局部信息融合的SVM多模态过程故障检测方法。运用局部概率密度方法对多模态数据进行预处理,消除多模态数据对工业过程故障检测特性的影响。在密度空间,分别运用主元分析(PCA)和局部保持投影(LPP)算法计算主元,提取数据的全局和局部信息,并将两者融合作为SVM的输入。运用正常和故障数据的全局和局部融合的信息训练SVM模型获得判别分类函数。建立模型之后,SVM能学习正常和故障数据的特性,从而将数据正确分类。将本方法运用于田纳西-伊斯曼多模态过程中,与传统PCA、LPP和SVM方法比较,实验结果进一步验证了本方法的有效性。

To effectively improve the fault detection performance of support vector machine(SVM)for multi-modal process,a SVM based on global and local information fusion for fault detection of multi-modal process method was proposed.The multi-modal data was preprocessed by local probability density method,and the influence of multi-modal data on the fault detection performance of industrial process was eliminated.In the density space,the principal component analysis(PCA)and local preserving projections(LPP)algorithms were used to calculate the principal components respectively.The global and local information of the data were extracted,and they were integrated as the input of SVM.SVM model was trained to obtain the discrimination classification function by using the information fusion of normal and fault data.After building the model,SVM learned the characteristics of normal and fault data to classify the data correctly.The proposed method is applied to the Tennessee-Eastman multi-modal process and compared with the traditional PCA,LPP and SVM methods.The experimental results further verify the effectiveness of the proposed method.