基于支持向量机的炭黑工艺建模  

Modeling Carbon-black Production Based on Support Vector Regression

作　　者：李梦龙[1] 刘军红[1] 黎金明[1] 陆敏春[1] 王克龙[1] 

机构地区：[1]四川大学化学学院,四川成都610064

出　　处：《应用基础与工程科学学报》2005年第1期51-57,共7页Journal of Basic Science and Engineering

基　　金：国家自然科学基金资助项目(No.29877016)

摘　　要：将支持向量用于炭黑工艺建模,并与主成分回归、反向传播人工神经网络以及径向基神经网络建模方法相比较.结果表明,炭黑生产过程具有比较强的非线性,不适合用主成分回归方法建立模型,支持向量机对炭黑吸碘值和吸油值的相对预测误差分别为1.62%和1.31%,所构建的模型的预测准确度明显优于反向传播人工神经网络(2.54%,1.64%),稍优于径向基神经网络(1.85%,1.38%).Support vector regression (SVR) is applied to the modeling of carbon-black production, and is compared with principal component regression (PCR), artificial neural network with error back-propagation (ANN-BP) and radial basis function network (RBFN). The results indicate that there is strong nonlinearity in the production of carbon-black, PCR is not suitable to make the model. The relative prediction error of SVR of iodine absorption value and DBP absorption value is 1.62% and 1.31%, respectively, which is obviously better than those of ANN-BP (2.54%, 1.64%) and slightly better than those of RBFN (1.85%, 1.38%).

关 键 词：支持向量机 工艺建模 炭黑 径向基神经网络 人工神经网络 反向传播 主成分回归 建模方法 生产过程 回归方法 非线性 吸油值 吸碘值 准确度 模型 预测 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP391.73[自动化与计算机技术—控制科学与工程]