Prediction Of Hearth Activity Based On MIC And Ridge Regression

As the core area of blast furnace,the activity of hearth directly affects the iron output and tapping quality of blast furnace.Quantitative analysis of hearth activity is of great significance for improving the service life of hearth and ensuring the stability and smooth operation of hearth.In view of the difficulties in obtaining blast furnace parameters,complex model calculations,and poor generalization ability in traditional quantitative model of hearth activity,this research starts from the process parameters collected from the blast furnace production site,and combines statistics,information theory and machine learning.A new prediction model of hearth activity is proposed,the specific contents are as follows:(1)Blast furnace data processing and analysis.For the process parameters collected from the blast furnace production site,first introduce the collection equipment such as blast furnace.Secondly,give a general description of the collected process parameters,and then perform data preprocessing on the collected data.This includes missing value filling,outlier processing,target value calculation,and data standardization.Finally,the evaluation index of regression model used in this paper is given.(2)Feature extraction method based on statistical correlation coefficient and information theory.For the treated data set,the Pearson correlation coefficient is selected from the statistical point of view to preliminarily select the characteristics of strong linear correlation with the hearth activity.Then,redundancy analysis is carried out on these features,and the redundant feature is deleted to get the feature subset 1.Then,in view of the defect that Pearson correlation coefficient can't capture the nonlinear relationship,the maximum information coefficient is used to measure the dependence between features and hearth activity from the perspective of information theory.After extracting features that are highly dependent on hearth activity,redundancy analysis is also performed.After deleting redundant features,feature subset 2 is obtained.(3)The prediction model of hearth activity based on ridge regression and support vector regression.For the constructed feature subset 1 and feature subset 2,first for the complex multi-collinearity existing in the feature subset,ridge regression is used to eliminate the effect of collinearity and predict the hearth activity.Since ridge regression cannot fit a nonlinear model,then support vector regression is used to establish the hearth activity prediction model.Finally,the four prediction models are compared and analyzed to find the optimal feature subset and regression prediction model.Through the evaluation index of regression model given in this paper,the four prediction models used in this study are compared.Among them,the best one is to extract features based on maximum information coefficient and establish prediction model by ridge regression.The R-squared value on the training set reaches 0.906,and the R-squared value on the test set reaches 0.897,which shows that the prediction model can well fit the blast furnace data.It also proves that the features extracted based on the maximum information coefficient have a significant impact on the hearth activity,and the ridge regression model can solve the multi-collinearity problem between features.From the perspective of relative error,the average relative error of the prediction model on the training set is 0.272%,and the test set is0.255%.This shows that the model does not have over-fitting problem,the model has strong generalization ability,and for new data the forecast effect is better.Finally,from the error distribution,the error between the predicted value and the actual value of the model is mainly concentrated in-5?5?,which shows that the model can accurately predict the hearth activity.Compared with the traditional quantitative model of hearth activity,the quantitative model of hearth activity proposed in this paper does not rely on the relevant experience in metallurgical field,the blast furnace parameters used in the model are easy to obtain,the number of parameters is small,the calculation is simple,and the model generalization ability is strong.