基于LTSA的FS-SVDD方法及其在化工过程监控中的应用  

FS-SVDD based on LTSA and its application to chemical process monitoring

作　　者：张少捷[1] 王振雷[1] 钱锋[1] 

机构地区：[1]华东理工大学化工过程先进控制与优化技术教育部重点实验室,上海200237

出　　处：《化工学报》2010年第8期1894-1900,共7页CIESC Journal

基　　金：国家杰出青年科学基金项目(60625302);国家高技术研究发展计划重点项目(2007AA041402);上海市重点学科建设项目(B504)~~

摘　　要：基于支持向量数据描述(SVDD)方法的非高斯过程监控和故障诊断具有众多优点。然而在对SVDD离线建模时需要在整个训练样本集上操作,对大样本集计算量相当大,也不利于在线操作时模型的更新。对此提出一种基于特征样本的SVDD(FS-SVDD),采用特征样本提取方法用少数几个特征样本代替原始数据集进行训练,显著降低了建模复杂度。同时,针对传统的线性降维算法如主成分分析(PCA)存在的提取过程数据非线性结构能力不足的缺点,首先用局部切空间排列(LTSA)方法提取出低维子流形,进行有效的维数约减;接着在这个低维子流形上执行SVDD算法;最后,利用相应统计指标进行过程监控。在TE过程上的仿真表明上述方法的有效性。Dozens of advantages have been reported on using support vector data description （SVDD） in the fields of non-Gaussian process monitoring and fault diagnosis.However,during the SVDD model construction offline,usually the whole training data set is used.Due to the tremendous size of the data set,the computation burden for modeling is quite heavy,which leads to the difficulty in updating model online.Therefore,this paper proposes a fast SVDD algorithm on the basis of the feature samples.In this new algorithm,the feature samples are used in stead of the whole training data set for modeling in order to significantly reduce the computation complexity.Concurrently,PCA is replaced by the local tangent space alignment （LTSA） to extract the underlying manifold structure of the process data set,since the traditional dimension reduction methods,such as PCA,have poor capability to handle nonlinearity.Next SVDD is applied on the manifold.At last,corresponding statistical indices are used for fault detection purpose.The proposed method has been tested on the Tennessee Eastman （TE） process,while the simulation results show the efficiency of it.

关 键 词：SVDD 特征样本 LTSA 过程监控 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]