基于DPCA-RBF网络的工业流化床乙烯气相聚合过程的软测量研究

工业流化床乙烯气相聚合反应是一个复杂的生产过程,具有高维、非线性、动态性和强噪声特点,质量变量难以直接测量。为解决关键质量变量在线软测量问题,首先采用动态主元分析(DPCA)的方法对过程变量提取主元,消除了过程变量之间的相关性、噪声并体现了建模数据的动态特性;其次对提取出的主元变量采用径向基函数网络(RBF)建模的方法,建立主元变量和质量变量之间的网络结构。对纯函数数据以及工业现场数据分别进行PCA-RBF模型及DPCA-RBF模型的仿真研究,研究结果表明,当建模数据存在非线性、动态性、噪声以及相关性等特性时,DPCA-RBF建模方法比PCA-RBF及单纯的RBF建模方法更优越。因此,DPCA-RBF建模方法较适合运用在工业实时变量的软测量中。

Soft-sensing research on the gas phase ethylene polymerization in fluidized bed reactor based on DPCA-RBF network

Gas phase ethylene polymerization in fluidized bed reactor is a complicated process,involving the characteristics of high dimension,nonlinear,dynamic and strong noise,and it's hard to measure the quality variables.To solve the problem of soft-sensing of the key quality variables on line,we first use DPCA to extract the principal component variables from the process variables aiming at eliminating noise and pertinence,reveal the dynamic identity between process variables,and then use RBF modeling method to establish the network topology between principal component variables and quality variables.We adopt PCA-RBF model and DPCA-RBF model on the function data and industrial real-time data respectively,and find that the modeling method of DPCA-RBF is better than the methods of PCA-RBF and RBF when nonlinear,dynamic,noise,and pertinence exist in the modeling data.Thereby,DPCA-RBF modeling method is more suitable for the soft-sensing of industrial real-time variables.