Improved Svm And Its Applications In Prediction Of Silicon Content In Hot Metal

Steel is one of the most important industrial raw materials for the development of modern society, while iron and steel industry is the pillar of modern state. In the steel industry's production process, the blast furnace ironmaking process is a very important part. In order to achieve energy saving and high yield stability and quality, a blast furnace process optimization and intelligent control system need to be built.Blast furnace process is a very complex physical and chemical reaction process of high temperature, including hundreds of physical, chemical reactions, involving hundreds of variables. The main variables include the speed of ore, the index of ventilating performance, the difference of iron, the quantity of blast, the temperature of blast, the quantity of coal powder and the rate of oxygen. One key technology of optimal control of blast furnace ironmaking is to accurately predict the change of the blast furnace temperature. Silicon content in hot metal ([Si])has long been used as one of the most important indices to represent the thermal state of the blast furnace, researchers have established a series of prediction model. According to the classification of mathematical models, these models can be divided into three categories:the mechanism model, the reasoning model and the mathematical model. With the deepening of prediction models, expert systems have been applied to the production of blast furnace, and achieved certain results. But the closed-loop intelligent control has not been achieved so far, still needs further study, of which the key technology is building an appropriate model to predict the temperature of the furnace.Statistical Learning Theory is a classic theory of machine learning, which follow the principle of structural risk minimization. The actual risk can be better optimized because the theory take into account both the empirical risk and the confidence interval. Support Vector Machine(SVM), a machine learning algorithm developed in statistical learning theory, can achieve the functions of classification and prediction. It is useful in solving the overfitting problem, while it can also obtain better learning results for the case of non-linear and limited sample. There are already some preliminary attempts to use the SVM in the prediction model of hot metal temperature. According to the characteristics of support vector machine, a fuzzy Support Vector Machine model and a mixed-kernel Support Vector Regression model are proposed and used in prediction of the silicon content in hot metal ([Si]). Analysis are made for 400 sample data from No.6 blast furnace(2500m3) of the Baotou Steel. Multiple experiments based on random sampling are carried out in order to avoid the influence of accidental factors. The results show that:compared with the decision tree and the traditional Support Vector Machine, the fuzzy Support Vector Machine model improve the classification results, raise the average hit rate of 10.3% and 2.2% respectively; compared with the ARM A model and the traditional Support Vector Regression, the mixed-kernel Support Vector Regression model improve the numerical prediction results, raise the average hit rate of 7.4% and 5.1% respectively.