改进仿射传播聚类的LS-SVM多模型建模方法  

An improved multi-model LS-SVM modeling method based on affinity propagation clustering

作　　者：熊伟丽[1,2] 张乾[2] 徐保国[1,2] 

机构地区：[1]江南大学轻工过程先进控制教育部重点实验室,江苏无锡214122 [2]江南大学物联网工程学院,江苏无锡214122

出　　处：《计算机与应用化学》2014年第7期802-806,共5页Computers and Applied Chemistry

基　　金：国家自然科学基金项目(21206053;21276111);中国博士后基金资助项目(2012M511198);江苏高校优势学科建设工程资助项目(PAPD)

摘　　要：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,但也存在着因模型选择失当而导致发生错误,为解决该问题,提出了一种新的建模方法。该方法先用仿射传播聚类算法实现数据聚类,并由最小二乘支持向量机建立各子模型。多模型预测时需知道待测样本对子模型的归属情况,则采用K近邻算法并结合隶属度阈值来进行判断。当待测样本对某子模型的隶属度大于阈值时,就由该子模型进行预测:若对所有子模型的隶属度均小于阈值时,则由K近邻算法从训练样本中选择与该待测样本相似的样本组成相似样本集,再采用最小二乘支持向量机建模并对该点预测。将其应用于青霉素发酵软测量建模中,并与其它方法比较,结果显示该方法是可行有效的,且能有效地克服当前一些多模型建模方法存在的不足。Comparing with single modeling method, multi-model modeling method usually own wider application range and better effect. But it is also easy to cause some errors due to improper model selection. In order to solve this problem, a new modeling method is proposed. First, an improved affinity propagation clustering algorithm is employed to decompose the data into several parts. Then least square support vector machine is applied to construct the sub-model. Multi-model based prediction needs to know the ownership situation of test samples, therefore K-nearest neighbor algorithm and the membership threshold value are adopted to judge it. A new sample＇s membership degree is greater than the threshold value of some sub-model, which it should belong to. Otherwise, the K-nearest neighbor algorithm is used to establish a similar sample set. Then least square support vector machine is adopted to build a new model. The effectiveness of the proposed approach is illustrated by applying to a soft sensor modeling of penicillin fermentation process and comparing with other methods. And this method can effectively overcome some shortcomings of other multi-model modeling methods.

关 键 词：仿射传播聚类算法 最小二乘支持向量机 K近邻算法 多模型 青霉素 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]