基于粗糙集-动态模糊神经网络的聚氯乙烯颗粒特性软测量

Based on Rough Set-Dynamic Fuzzy Neural Network Soft Sensor of Particle Properties of Polyvinyl Chloride

提出了一种基于粗糙集-动态模糊神经网络的软测量建模方法,并将其应用于聚氯乙烯平均粒径软测量研究中。首先,用粗糙集隶属函数对各个辅助变量进行分析比较,得到最简的辅助变量个数,然后采用等间距离散法对各个辅助变量和主导变量进行离散化,分析得到离散表和最简模糊规则数,并采用模拟退火法对动态模糊神经网络输出层的权值参数进行优化训练。研究结果表明,与常规动态模糊神经网络相比,基于粗糙集-动态模糊神经网络的聚氯乙烯平均粒径软测量模型具有更佳的预测效果。

Rough set-dynamic fuzzy neural network soft sensor is proposed,and used to predict average particle diameter of polyvinyl chloride.The rough set membership function is used to reduce the number of auxiliary variables.The discrete table and the simplest number of fuzzy rule are obtained through discretion of every auxiliary variables and dominant variable using the equidistant discrete method.The weight parameters of dynamic fuzzy neural networks output layer are optimized and trained using simulated annealing method.The results of using rough set-dynamic fuzzy neural network model in research of the average particle diameter of polyvinyl chloride have better prediction,as the conventional rough set-dynamic fuzzy neural network model.