Applications Of SVM To Predict Silicon Content In Hot Metal

Blast Furnace ironmaking, which is the main working procedure of the steel industry, is the pillar of the national economy, and is very important for the steel industry and economizing energy sources. B.F ironmaking process is highly complicated, whose operating mechanism is characteristic of nonlinearity, time lag, high dimension, big noise and distribution parameter etc. The models to predict and control the silicon content in molten iron is the kernel of automation in B.F ironmaking. Exact prediction of the silicon content can help foreman to enhance operation level and retrench coke rate.SVM(Support vector machines) is a novel arithmetic of machine learning which is based on SLT(StatisticaI Learning Theory).It can solve the problem characterized by nonlinear, high dimension, small sample and local minimizing perfectly. Presently it has been widely used in pattern recognition, function approximation and data mine etc. In this paper, a new approach to predict the silicon content in molten iron (the silicon content in molten iron reflects the chemistry temperature of B.F, might be used to express the temperature of B.F) and classifier of silicon content in molten iron based on SVM are proposed. On the basis of analyzing the data from the Intelligent Control Expert System on 750m~3 BF Laiwu Iron & Steel Group Co. and ironmaking knowledge, SVM model is established to predict the silicon content in molten iron.There are four parts presented in this paper. In the first part, the B.F ironmaking process, B.F Expert System and the methods to predict the silicon content in molten iron are introduced. In the second part, the control parameters and state parameters of B.F ironmaking process are analyzed. In the third part, the Statistical Learning Theory and Support Vector Machine are introduced. In the fourth part, the model to predict the silicon content in molten iron and classifier of silicon content in molten iron based on SVM are proposed.Time series data of [Si] in hot metal actually observed in No.1 blast furnace at Laiwu Iron and Steel Group Co. are taken as sample space with the capacity of 1000 tap numbers and sampling time interval of 2h. The data of [Si] is divided into 5 classes according to C-mean clustering algorithm. In this paper, multi-class classification of [Si] by improved M-ary classification is proposed, and abilities of different classifiers are compared. The prediction method based on SVM get a higher hit ratio, which come up to more than 85%, than those of AR model, which only can reach 75%.