竞争聚类的邻近支持向量机软测量研究应用  

Research and Application of Soft Measure based on Proximal Support Vector Machine with Competitive Clustering

作　　者：王华秋[1] 

机构地区：[1]重庆理工大学计算机学院,重庆400054

出　　处：《控制工程》2013年第4期654-658,662,共6页Control Engineering of China

基　　金：教育部人文社会科学研究青年基金项目(10YJC870037);重庆市教委科学技术研究项目(KJ100805)

摘　　要：提出了一种基于竞争学习聚类的邻近支持向量机的软测量方法。本文用数据点的密度来近似衡量数据点相对于类中心的位置,从而改进了竞争学习聚类算法的权值调整方法。采用改进的竞争学习聚类算法将训练样本聚集到不同的中心,采用邻近支持向量机对每一类进行训练建立子模型,利用样本方差进行计算邻近支持向量机的惩罚因子,并计算新增样本数据和所有类的相似度,对被检索出的相似子模型参数加权,计算预测结果,用新增加的样本更新训练数据及其聚类中心。将所提出的方法用于氧化铝拜尔溶出过程关键化验量的软测量,解决了模型失效问题,实验表明:该方法有效地增强了软测量模型适应工况变化的能力,提高了预测精度。A proximal support vector machine with competitive learning clustering is proposed for soft-measuring. In this paper, the density of data points is used to approximately measure the relative location of data points to the cluster centers, thus the weight adjust- ment method of competitive learning clustering algorithm is improved. The improved competitive learning clustering algorithm is used to cluster the training samples to different centers and each clustering is trained to establish sub-model by proximal support vector machine （PSVM）. Sample variance is used to calculate the penalty factor of PSVM. Then the similarities of increasing sample data and all clus- ter centers are calculated, the parameters of the retrieved similar sub-model is weighted to output the prediction result. The increased sample is used to update training data and their cluster centers. The proposed method has been used into the soft measure of key chemical parameter in the process of Bayer dissolution of alumina to solve the failure problems of soft measure model. The experiments show that the adaptive capacity of soft measure model has been enhanced by the method and its predicting accuracy is increased.

关 键 词：软测量 竞争学习聚类 邻近支持向量机 拜尔溶出过程 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]