Prediction Of Hearth Activity Based On Data Mining Methods

The activity of the hearth is like the heart of blast furnace production.Once hearth activity is broken,stable operation of the blast furnace will be destroyed,which will bring huge economic losses to the enterprise.At present,the quantitative calculation of hearth activity is only carried out from the perspective of empirical formulas,and it is difficult to obtain parameters.Besides,the early warning function of hearth activity can not be achieved.In response to this problem,this study proposes data mining methods to predict activity of the blast furnace hearth by using operating parameters in the actual operation of the blast furnace.This includes:(1)Data processing and model evaluation criteria.Firstly,the method and way of data acquisition are briefly introduced.Besides,the acquired data is simply analyzed.Secondly,data cleaning operations such as data filling and removing are carried out,and the target value is calculated.Besides,the processed data is standardized.More importantly,the data set is divided.Finally,the evaluation criteria of the model in this paper are given.(2)Prediction of hearth activity based on the improved multiple linear regression model.For the processed data,correlation degree between operation parameters and the target value is first obtained by using correlation analysis,and the strongly correlated operation parameters are selected as preliminary features.Then the relationship of physical formula between initial features is investigated to eliminate redundant characteristics.Finally,the Akaike Information Criterion is introduced as loss function to establish models for predicting hearth activity.(3)Prediction of hearth activity based on the regression model of support vector machine via principal component analysis.The multiple linear regression model only reflects the linear relationship between operation parameters and the target value.In order to explore the nonlinear relationship between operation parameters and the target value,the principal component analysis algorithm is used to convert the original data into independent components while ensuring that the loss of the original data is as small as possible.Then the component is used as the input of the support vector machine regression model to establish support vector machine regression models based on different kernel functions to predict hearth activity.Through the analysis of above methods,for the prediction of hearth activity,the improved multiple linear regression model and support vector machine regression model have obtained good prediction results.Average relative errors of the improved multiple linear regression model and the optimal support vector machine regression model on the test set are 0.69% and 0.44%,respectively.And average relative errors on the training set are 0.57% and 0.42%,respectively.Besides,both models have the early warning function.Compared with the multiple linear regression model,prediction of hearth activity based on the optimal support vector machine regression model via principal component analysis reduces the average relative error on the test set by 36.23% and has stronger early warning ability,which reflects the nonlinear relationship between original data and the target value to a certain extent.Compared with the other work,the quantification model of hearth activity proposed in this paper has the advantages of independent of experience,easy acquisition of parameters and early warning function.