中文摘要：

将主元分析（principal component analysis,PCA）模型相似度（以下简称PCA相似度）和谱聚类（spectral clustering）算法相结合,并用于基于高炉历史数据挖掘的炉况工作点变化的分析。利用PCA相似度与距离相似度的加权来衡量滑窗数据集之间的相似度,进一步将数据集的聚类问题转化为图的最优划分问题,通过谱聚类得到聚类结果。该方法降低了高炉工作点漂移的影响,能够有效稳定的实现高炉炉况工作点的聚类。基于现场历史数据的离线测试表明：与已有的基于PCA相似度和k-means聚类的算法对比,本研究可以更加有效区分炉况工作点的跳变。

英文摘要：

The principal component analysis （PCA） similarity factor and spectral clustering algorithms were combined and applied analyze the operational state change in a blast furnace by mining the historical data. The similarity between different data sets generated from moving windows by combining the PCA similarity factor and the distance similarity factor was measured, and the historical data were clustered by constructing the graph from the similarity between different data sets and using spectral clustering algorithm. The effect of operating point drift was reduced and the more accurate clustering result was effectively and steadily achieved by the proposed method. The off-fine test proved that, compared with the existing methods which combined the PCA similarity factor and k-means clustering, the proposed method could more effectively recognize the operational state change in a blast furnace.