| Blast Furnace ironmaking, which is the main working procedure of the steel industry, is the pillar of the national economy. And it is very important to the development of steel industry. During the actual production of the Blast Furnace, if the fault happens, it at least impact on the amount of iron, and worst of all, it will make a great loss to the steel factory. So the research on fault diagnosis of Blast Furnace means a lot to the operation of Blast Furnace.There are hanging、slipping、blast furnace wall liner thickening、cooling and warming as the common abnormal Blast Furnace condition. The fault of Blast Furnace has an obvious feature, which is the kind of problem of small sample. In response to this problem, we take the method of combining one-against-rest SVM with RBF neural network. With this method, we both solve the problem of the small sample and the problem which may happens during the classification of one-against-rest SVM to improve the accuracy of fault diagnosis.In the theory of SVM, penalty factor C and kernel function’s factor plays an important role in classification accuracy. So we take the method of PSO to optimize the factors to ensure the accuracy of diagnosis. On this basis, in response to the sample skew problem, we use the hypersphere theory to optimize the penalty factor C of the training sample. With this method, the accuracy has been proved again.In order to verify the effectiveness and feasibility of the method, we studied in Matlab. First, we use the IRIS data of UCI to verify the feasibility of the method. And then, after combining the experience of site operators and the content of reference, we choose the fault parameter and pretreat the data obtained from steel factory. On this basis, we take the method above to carry out the fault diagnosis of cooling and warming of Blast Furnace. The result shows that, the method to combine SVM with RBF can make an accurate diagnosis on such fault and it is better than using the method of SVM alone. |
PDF Full Text Request