一种聚类加权支持向量机算法及其在软测量中的应用

A clustering weighted SVM algorithm and its application in soft-sensing

针对支持向量机应用于软测量建模时,工业过程数据中特异点影响建模精度的问题,提出聚类加权支持向量机方法。该方法首先对建模数据进行聚类分析,根据聚类结果,对各类数据的惩罚系数进行相应的加权,改变权值大小既能减小特异点对模型的影响程度,又能将其包含的生产过程信息引入到软测量模型中。聚丙烯熔融指数软测量的实例研究表明,通过对建模数据进行聚类分析和加权处理,聚类加权支持向量机比标准支持向量机建模更准确。

Some special points were included in industrial collected data. When used for soft-sensing, the special points can influence the modeling performance. A clustering weighted SVM method was proposed on consideration of the special points. The approach first applied cluster analysis to those process data so as to distinguish the normal data from the special points. According to the cluster results, different weighting factors were assigned to the penalty parameters of different subsets respectively. The influence of the modeling data on model can be adjusted to improve the accuracy of the soft-sensing by changing the weighting factors. The simulation results on polypropylene MI demonstrate that the clustering weighted SVM is more accurate than SVM in soft-sensing modeling.