基于核熵成分分析的热轧带钢自适应聚类分析

为提高热轧带钢力学性能离线检测的针对性和生产过程控制的实时性,提出利用聚类分析方法实现生产状态的聚类,对错分或离群样本进行力学性能的重点检测.常用的高斯核主成分聚类分析中假设数据服从正态分布,以方差大小提取核主成分,而实际生产数据分布复杂,拟采用核熵主成分分析,并自适应选取核参数和聚类数,实现生产状态的自适应聚类.利用实际生产数据进行方法验证,与核主成分聚类分析相比具有更好的聚类结果,聚类正确率从86.23％提高到96.51％,更加有效地提高了质量检测的针对性.

Hot rolled strip state clustering based on kernel entropy component analysis

In order to detect mechanical properties of hot rolled steel offline more efficiently and to the point,and improve the control timely,the process data are used for clustering analysis to acquire the state in advance.During the mechanical properties detection,the outliers in the same steel grade are regarded as the focal points.On the assumption that the data obey normal distribution,the variance is used as the information metric to extract features in the common method kernel principal component analysis(KPCA) with Gaussian kernel.The actual production data have complex distribution.Then,kernel entropy component analysis(KECA) was used to extract features and the kernel parameter and cluster number were selected adaptively to do the production state clustering.The real hot strip rolling process data are used for model validation,and as a result,the proposed method has better performance on clustering compared with KPCA to enhance the pertinence of quality detection.The clustering accuracy is improved from 86.23% to 96.51%.