关键词: 化工系统, 海量数据, 故障分类, 流形学习, 扩散映射, 支持向量机, 故障诊断

Abstract: To distinguish the fault type by extracting fault features in chemical process system,a novel fault classification method named Diffusion Mapping-Support Vector Machines(DM-SVMs)based on mass data was proposed.Through the diffusion mapping and its linear incremental approach,the online data of chemical system in high-dimension space was mapped to low-dimension subspace,and the manifold features were classified by multiclass support vector machine.The proposed DM-SVMs approach was used to train the multiple fault classification which could realize the anomaly identification of support vector machine.Through verifying the data of Tennessee-Eastman chemical process and realistic chemical process,the feasibility and efficiency of proposed method were proved.Comparing to other Manifold-SVMs methods,the DM-SVMs was proved to be more precise.

Key words: chemical system;mass data, fault classification, manifold learning, diffusion features, support vector machine, failure diagnosis