基于反应基元的非线性系统灰箱建模方法

Grey-box modeling based on fundamental genes for nonlinear system

提出一种基于反应基元的建立复杂非线性系统模型的灰箱建模方法。首先根据先验知识及系统特性分析引入过程的初始反应基元,并以此为出发点建立结构逼近神经网络模型,实现基元之间的关联,赋予网络节点实际的物理意义;然后,通过提出的最小化预测误差,结合逐步回归分析方法选择最优反应基元,优化网络结构,建立起表示系统变量关系的灰箱模型。以实际橡胶硫化促进剂制备的间歇反应过程作为实验对象,建立以生成物浓度为输出的数学模型,达到较高的输出预测精度。

An approach of grey-box modeling based on fundamental genes was developed for modeling dynamic processes with non-linear characteristics.By combination the prior knowledge and systematic behaviors,structure approaching neural network（SANN） was established based on fundamental genes,and the nodes of SANN were given actual significance.Then the optimal fundamental genes were chosen through minimizing the proposed predicted error with stepwise regression analysis（SRA） to optimize the structure of SANN,so as to get the grey-box model.Detailed process of modeling was described in modeling of batch condensation reaction of producing promoter for vulcanizing rubber.The simulation result proves that the approach is effective.