基于改进自适应仿射传播聚类的多模型建模研究  

Multiple models soft-sensing method based on improved adapt affinity propagation

作　　者：付晓幸 于佐军[1] 

机构地区：[1]中国石油大学(华东)信息科学与工程学院,山东青岛266580

出　　处：《计算机与应用化学》2016年第1期111-116,共6页Computers and Applied Chemistry

摘　　要：聚丙烯装置熔融指数软测量中存在非线性和多工况切换操作的问题,针对普通的自适应仿射传播聚类查找最优聚类耗时较长的问题,提出一种改进的自适应仿射传播聚类,通过计算偏向参数范围缩小搜索空间,提高聚类速度和精度。多模型建模方法通常比单一模型建模方法适用范围更广、效果更佳,小波核函数不仅具有非线性映射的特征而且也具有小波分析对非平稳信号的逐级精细描述的特征,本文提出将小波核函数与正交最小二乘法相结合的方法分别对数据子集建立模型,粒子群算法实现对参数的选择,能够以较高的精度逼近函数,并通过开关切换方式根据当前工作点所属子类模型进行预测输出。通过对聚丙烯熔融指数的软测量建模研究表明,本文提出的方法具有良好的回归精度和较好的泛化性能。Modeling polypropylene melt index soft senor is a progress that nonlinear and multi-grade. The standard affinity propagation clustering algorithm suffers from one limitation that it is hard to know the value of the parameter preference which can yield an optimal clustering solution. And it takes long time to find the optimal clustering number for ordinary adaptive affinity propagation clustering. To overcome this limitation, in this paper we proposes an improved adaptive affinity propagation clustering. The method first finds out the range of preference, then search the space of preference to find a good value which can optimize the clustering result. Compared with single model, Multi-model modeling methods usually have better applicable and effect. Wavelet kernel function not only has the characteristics of nonlinear mapping, but also has the characteristics of the nonstationary signal by wavelet analysis. Combined the wavelet kernel function of particle swarm optimization and the orthogonal least square method to build the model of data set, it is able to approximate function with higher accuracy. The test samples are assigned to appropriate sub-class, then predicted outputs are estimated by corresponding sub-models. The simulation results of Melt Index indicate that the proposed method have better prediction accuracy and generalization performance.

关 键 词：软测量:多模型 仿射传播聚类算法 正交最小二乘法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]