基于小波去噪核主元分析和邻近支持向量机的性能监控和故障诊断  

Performance Monitoring and Fault Identification Based on Denoised Kernel Principal Component Analysis and Proximal Support Vector Machine

作　　者：张曦[1] 阎威武[1] 赵旭[1] 邵惠鹤[1] 

机构地区：[1]上海交通大学自动化系,上海200240

出　　处：《上海交通大学学报》2008年第2期181-185,共5页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金(60504033);浙江大学工业控制技术国家重点实验室开放课题(0708004)资助项目

摘　　要：针对化工过程数据中包含噪声和强非线性的特点,提出了基于小波去噪核主元分析(De-noised Kernel Principal Component Analysis,DKPCA)和邻近支持向量机(Proximal Support Vector Machine,PSVM)的性能监控和故障诊断新方法.将样本数据用小波方法进行去噪处理,去除数据所包含的噪声,通过KPCA将降噪后的数据进行变换,在特征空间里构建T2和Q统计量来监测是否有故障发生;若发生故障,则计算数据的非线性主元得分向量,并将其作为PSVM的输入值,通过PSVM分类来确定故障的具体类型.流化催化裂化装置(FCCU)仿真试验验证了小波去噪的必要性和利用DKPCA-PSVM进行监控和故障诊断的有效性.The data collected from chemical process are strongly nonlinear and include noises, which influences the performance of traditional SPC methods. To solve this problem, a novel combined method for performance monitoring and fault identification using denoised kernel principal component analysis（KPCA） and proximal support vector machines （PSVMs） was proposed. It first removes the noise from data set using wavelet transform. Then the denoised data is disposed using KPCA and T^2 and Q statistics are con- strueted in the feature space. If the T^2 and Q statistics exceed the predefined control limit, a fault may have occurred. Then the nonlinear score vectors are caleulated and fed into the PSVMs to identify the faults. Fluid catalytic cracking unit（FCCU） simulator proves that the DKPCA-PSVM method for process monitoring and fault identification is effective and the denoising using wavelet transform is essential.

关 键 词：小波去噪 性能监控 故障诊断 小波变换 核主元分析 邻近支持向量机 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]