ε不敏感支持向量回归在化工数据建模中的应用  

Chemical process data modeling based on ε-insensitive support vector regression

作　　者：邵信光[1] 杨慧中[1] 石晨曦[1] 

机构地区：[1]江南大学控制科学与工程研究中心,无锡214036

出　　处：《东南大学学报（自然科学版）》2004年第B11期215-218,共4页Journal of Southeast University：Natural Science Edition

基　　金：国家高技术研究发展计划 (863计划 )资助项目 (2 0 0 2AA412 12 0 ) .

摘　　要：针对化工领域数据建模小样本、不适定性等问题 ,提出了一种用ε不敏感支持向量回归 (ε SVR)方法进行实际过程建模的想法 ,以解决人工神经网络等方法在数据建模中的“过拟合”、泛化性差等问题 .在分析ε SVR特性的基础上 ,用一个非线性函数逼近例子验证了ε SVR在小样本情况下比BP前馈神经网络具有更优良的建模能力 .将ε SVR应用到丙烯腈聚合反应过程质量指标软测量混合模型中 ,仿真和现场运行结果表明ε SVR是一种非常有效的化工数据建模方法 .Since the size of the sample data is small and the sample is ill-posed in many chemical industry processes, to avoid the problem of overfitting and poor gen eralization capability in conventional data modeling methods (such as back propa gation feedforward neural networks), a new way of support vector machines for da ta modeling is presented. Based on the analysis of ε-insensitive support v ector regression (SVR), including details of the algorithm and its implementatio n, the advantages of ε-SVR were demonstrated through a nonlinear function approximation example where the size of the sample data is small. Furthermore, t he ε-SVR method was successfully applied to a hybrid model for the perform ance figure of polyacrylonitrile productive process, and the simulation results show that the ε-SVR is effective for chemical process data modeling .

关 键 词：ε不敏感支持向量回归 聚丙烯腈 软测量 数据建模 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]