Fault Diagnosis of Blast Furnace Based on DAGSVM

For achieving high efficiency of artificial intelligence applied in decision-making system of blast furnace, and reducing high technique demands to operators, a new multi-classification method based on support vector machines (SVMs) is proposed. In order to avoid dimension disaster and solve multi-classification problem, use decision directed acyclic graph (DDAG) algorithm combined with each kernel function, and map the training samples into high dimension space utilizing the statistic learning theory. Then compare different performance of each kernel function referring to the actual process data, and select the proper one to construct diagnosis classifier. Through tested different multi-classification strategies, simulation results show that DAGSVM model is superior to the others on testing accuracy and has efficient classification ability