Abstract: Taking into consideration the low efficiency of applying intelligence to blast furnace fault diagnosis and the high demand to operator's technique,a multi-classification method based on support vector machine(SVM) is proposed.According to statistic learning theory,we use kernel functions to map the training samples into a high dimensional space for training.Combining the testing accuracy of different kernel functions,an optimal kernel function is obtained to solve this problem.By comparing different multi-calssification strategies,a diagnosis model based on DAGSVM(directed acyclic graph SVM) is constructed.Experiment results show that the proposed algorithm has a higher identification accuracy.