基于小波-最小二乘支持向量机的熔融指数预测  

Prediction of Melt Index Based on Wavelet and LS-SVM

作　　者：刘美[1,2] 陈政石[1] 孙宗海[2] 

机构地区：[1]茂名学院自动化系,广东茂名525000 [2]华南理工大学自动化科学与工程学院,广东广州510640

出　　处：《计算机仿真》2009年第10期162-165,共4页Computer Simulation

基　　金：国家自然科学基金项目(60704012);广东省自然科学基金项目(06300232);广东省科技计划项目(2004B10301035);茂名市科技计划项目(201095)

摘　　要：熔融指数是聚丙烯生产的主要质量指标,为了在线监测熔融指数,并实现质量的自动监控是聚丙烯生产工业亟待解决的问题。对液相间歇式聚丙烯反应机理进行分析,找出影响熔融指数的主要参数,采用小波基函数对原始数据进行粗大误差处理,主成份分析确定最少辅助变量,简化支持向量机结构,应用交叉验证法优化最小二乘支持向量机模型参数,建立熔融指数预测模型。仿真结果表明,熔融指数的预测平均绝对误差为1.923,均方差为1.219。预测精度满足生产要求,且模型泛化能力强,运行速度快,可以有效指导生产。Melt index is a dominate quality index of polypropylene. Monitoring the melt index on - line, and realizing automatic quality control of production process are urgent problems in the production of polypropylene. In this paper, several major parameters affecting Melt index were firstly selected based on mechanism analysis. Then wavelet functions were used to remove error data. Principal Components were extracted by PCA and used as secondary varia- bles of LS - SVM. Finally the optimum parameters of LS - SVM were selected by cross validation and a prediction model for the Melt index was established. The simulation results show that, the average absolute error and MSE are 1. 923 and 1. 219 respectively. The model features high accuracy, good approximation and generalization ability. This model may be used to efficiently guide production.

关 键 词：熔融指数 预测 最小二乘支持向量机 主成份分析 小波 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]