基于t分布随机邻域嵌入算法的工业过程故障分类

Industrial process fault classification based on t-distributed random neighborhood embedding algorithm

针对在工业过程中数据普遍存在的非线性特性,基于数据的局部相关关系对分类的影响,提出一种基于t分布随机邻域嵌入(t-SNE)的数据特征提取和故障分类方法。利用t-SNE算法非线性、非参数降维的优势,与费舍判别分析(FDA)、支持向量机(SVM)分类器相结合建立故障分类模型。利用t-SNE算法对故障数据进行非线性特征提取,获取数据的关键区分特征。用FDA和SVM算法实现故障分类和识别。通过田纳西-伊士曼(TE)过程获得的实验数据进行实验仿真分析,并分别与基于核主元分析法(KPCA)、拉普拉斯特征映射(LE)构建的KPCA-FDA、LE-FDA、KPCA-SVM、LE-SVM 4种故障分类模型进行比较。定量评估结果表明:即使基于不同分类器,相较于其他2种方法,该文方法的分类准确率分别提升了2%和7%,且其平均分类准确率能保持在97%以上。

Aiming at the nonlinear characteristics of data in industrial process,and based on the influence of local correlation of data on classification,a data feature extraction and fault classification method based on t-distributed stochastic neighborhood embedding(t-SNE)is proposed.The method makes full use of the advantages of nonlinear and non-parametric dimension reduction of t-SNE algorithm,and combines with Fisher discriminant analysis(FDA)or support vector machines(SVM)classifier to establish fault classification models.The t-SNE algorithm is used to extract the nonlinear features of the fault data,and the key distinguishing features of the data are obtained.The FDA and SVM algorithms are used to classify and identify faults.The experimental simulation analysis is carried out by Tenessee Eastman(TE)process,and is compared with the KPCA-FDA,LE-FDA,KPCA-SVM,LE-SVM four fault classification based on the kernel principal component analysis(KPCA)and Laplace eigenmap(LE).The quantitative evaluation results show that:even based on different classifiers,compared with the other two methods,the classification accuracy of the proposed method is improved by 2%and 7%respectively,and the average classification accuracy can be maintained above 97%.