Study For Prediction Model Of Blast Furnace Temperture Based On Principal Component Analysis And Partial Least Squares

Blast furnace (BF) iron-making production process was a very complicated and highly coupled nonlinear system in actural production. Nonlinear, time-varying, higher dimensionality and bigger noises were the notable characteristics of its operating mechanism. In blast furnace iron-making process the blast furnace temperature control was a key and important factor for blast furnace steady and high efficiency production. But dozens of important parameters were affected the blast furnace temperature control and the impacts also existed among the various parameters. In order to control the number of parameters, excavate the potential factors on blast furnace temperature and improve the accuracy of blast furnace temperature prediction model, it was very effective to use principal component analysis (PCA) method and the partial least squares (PLS) method.Lots of primitive data from Baotou Steel No.6BF on-line collection were used to calculate and analysis in this paper. Most of the parameters affecting blast furnace temperature had certain time delay. So firstly influence parameters of blast furnace temperature were analyzed time delay based on Shannon entropy of generalized correlation coefficient method.And secondly four principal components based on total variance contribution with 87.72% were extracted according to principal component analysis from 16 variables that impacting silicon content. A regression model for predictive control of silicon content was established. The model was tested on datasets from No.6 BF in Baotou Steel Corporation and good accuracy was received with 88.4%. According to the model air volume and air temperature control range were gave on the influence of molten iron blast furnace silicon content. Meanwhile, single dependent variable partial least squares was applied for modeling to use 8 variables of the above 16 variables, extracting three components explain 88.83% variation of information, and accuracy of 89.2%. From the comparison of two models, we can see that blast furnace temperature prediction model based on partial least-square regression method has more advantages of simple calculation, high precision accuracy and easy qualitative explanation than that based on principal component analysis. Finally blast furnace status forecast model was established using the method of multivariable partial least-square in this paper. Using multivariable partial least-square method modeling thought to establish the partial least-squares regression model based on dependent variables of silicon content, deoxidization rate of iron, sulfur content, slag ratio and independent variables of eight indexes from the above sixteen indexes. Good forecast was achieved in blast furnace status. So we can grasp better blast furnace status condition and the production process with this theory basic.