论文部分内容阅读
针对可拓神经网络(ENN)对复杂样本数据分类效果较差的问题,提出一种融合边界判别投影(MDP)和改进半监督近邻传播(ISAP)的新型ENN分类算法。首先,使用边界判别投影对原始数据进行降维,提取关键特征。其次,在低维特征空间进行聚类分析,利用近邻传播聚类筛选出有效训练样本,并通过ISAP聚类寻优获得样本类中心,作为初始类中心,在此基础上构建新的分类器。最后,将其应用于复杂化工过程中高密度聚乙烯(HDPE)的熔融指数预测,取得了较好效果。
Aiming at the poor classification effect of ENN on complex sample data, this paper proposes a new ENN classification algorithm that combines boundary discriminant projection (MDP) and improved semi-supervised neighbor propagation (ISAP). First, using the boundary discriminant projection to reduce the dimension of the original data and extract the key features. Secondly, the clustering analysis is carried out in the low-dimensional feature space, and the effective training samples are screened by the neighborhood propagation clustering. The sample centers are obtained by ISAP clustering as the initial class centers, and a new classifier is constructed based on the clustering analysis. Finally, it is applied to the prediction of melt index of high density polyethylene (HDPE) in complex chemical process, and achieved good results.