Abnormal Detection of Blast Furnace Condition Using PCA Similarity and Spectral Clustering

Aiming at the fast-changing operating points challenge in blast furnace, a method for operating point identification of hot-gas stove exchange combining principal component analysis (PCA) similarity and spectral clustering is proposed in this paper. Firstly the NJW spectral clustering algorithm based on weight PCA similarity fusion is used to cluster the working modalities and identify the operating points for early blast furnace abnormal condition detection. Then, the convex hull is introduced to identify the hopping point of hot blast furnace and distinguish the hot blast furnace from the abnormal furnace condition so as to further reduce the blast furnace abnormality false alarm rate. Experimental Results demonstrate that the proposed method can effectively reduce the false alarm rate and reduce the impact of training set selection process on the detection results of early blast furnace abnormal condition detection based on PCA multi monitoring.