Fault diagnosis of blast furnace based on improved binary-tree SVMS

Support vector machine (SVMs) is powerful for classification problem with small sampling, nonlinear and high dimension. In this paper, two new improvements of SVMs algorithm on samples pretreatment and SVMs binary-tree construction are proposed to solve fault diagnosis problem of blast furnace in ironmaking. The input data of diagnosis system is preprocessed through a special method based on reducing useless samples technology and some characters are extracted for according to them diagnosing blast furnace faults. A new improved binary tree SVMs multi-class classification algorithm is proposed and applied to diagnosis of blast furnace. The experiment results show that the improved binary-tree SVMs algorithm has an excellent performance on training speed and diagnosis accuracy.