Data-driven Modeling And Prediction Of Hot Metal Silicon Content

The development of steel industry evaluates the performance of national economy,and its output and quality are of great significance to the construction of national economy.As the material basis of the subsequent steelmaking,the quality of molten iron is extremely important to steel production.In the blast furnace ironmaking process,the silicon content of molten iron is an important factor for characterizing the internal thermal state of blast furnace and an important indicator for measuring the quality of hot metal as well.Therefore,modeling and prediction of hot metal silicon content establish a solid foundation for the high performance of operation control during the blast furnace ironmaking process and guarantee the stable operation of ironmaking blast furnace in the long term.For the modeling and prediction of the hot metal silicon content with single variable time series,we first decomposes the time series of the hot metal silicon content into several non-interfering and non-intersecting intrinsic mode function(IMF)and a residue(Rn)by empirical mode decomposition(EMD).Then we approximate each IMF and Rn to Nonlinear Autoregressive Model and make the prediction using support vector regression(SVR)and dynamic neural network(DNN)respectively and combine the separate prediction result to finally get the prediction of silicon content.At last,we perform some experiments on some sample data collected from No.2 blast furnace in Liuzhou(LZ)ironwork.The experiment results show that the proposed combinational algorithm of EMD_SVR and EMD_DNN precedes the traditional non-EMD algorithm in terms of both mean square error(MSE)and hit rate.For the modeling and prediction of hot metal silicon content with multivariable time series,we first select the input varibles for the prediction algorithm of hot metal silicon content from 20 factors like hot blast pressure,cold air flow and oxygen enrichment that collected from No.2 blast furnace in LZ ironwork by combining the mechanism of blast furnace ironmaking with principal component analysis.Furthermore,the lag time of each input variable in the prediction algorithm is determined by combining the maximum information coefficient(MIC)with the experience of blast furnace ironmaking process.Based on the input variables and lag-time parameters,SVR and DNN are used to establish nonlinear autoregressive models with exogenous inputs(NARX)for the prediction algorithm of hot metal silicon content.At last,we verify the algorithm with experiments using some sample data that collected from No.2 blast furnace in LZ ironwork.The experiment results show that the methods proposed in this paper have their own merits.And they are superior to other traditional algorithms in terms of time,MSE and hit rate which verifies the feasibility and innovation of the data-driven modeling method that we proposed.The prediction algorithm of the hot metal silicon content proposed in this paper is verified by the experiment with actual data.Given full consideration of the need from the ironmaking process for the No.2 blast furnace in LZ ironwork,we design and implement a metal quality forecasting system using LABVIEW and MATLAB.The system mainly consists of PC display,server and database and realizes some functions including system log-in,silicon content of hot iron,blast furnace operation status monitoring as well as data recording and preservation.