Data Mining And Application Of Sintering Production Process

The sintering process is a strongly coupled,multivariable,nonlinear dynamic system.The sintering process is of great significance for the smelting of blast furnaces.The content of sinter less than 10 mm and the content of FeO are two important indicators to measure the productivity quality of sinter.The quality of these two indicators directly affects the output and quality of the sinter,which in turn affects the economic benefits of steel companies.The time-varying characteristics of the sintering process make it difficult to describe and control the fluctuations of these two indicators using the traditional mechanism model.Therefore,the control of sinter quality indicators has always been an intractable problem.With the advancement and improvement of big data technology,the thinking of big data has slowly penetrated into all walks of life.Major enterprises have established the idea of big data modeling and prediction based on sintering production process,using cutting-edge integrated learning and deep learning algorithms.Combined with the sintering production process,the content of sintered ore less than 10 mm and the content of sinter FeO were modeled.The sinter plant database has accumulated data of various types and different structural forms for many years.In order to extract useful information from these massive data,it is necessary to collect and sort the raw material process,the mixing process,the production operation equipment,and the sinter quality inspection process.data.Identify the outliers of the data using box plots and eliminate them in conjunction with process requirements,the null values are interpolated using the median method to create a database that meets the needs of the predictive model.Secondly,the characteristic engineering is carried out for two prediction models of sintered ore with less than 10 mm grain size and sinter FeO content.The xgboost algorithm is used to sort the features that affect the content of sinter or less than 10 mm and to screen the features.The random forest algorithm is used to sort the characteristics of the FeO content of the sinter and to screen the features.The heatmap is used to visualize the correlation of features,and the features with less influence on the target parameters are further eliminated.The 14 features that have a greater influence on the content of sinter less than 10 mm are selected,and 45 features that have a greater impact on the FeO content are also screened.At the same time,exploratory factor analysis was carried out on some features.Thirdly,a deep neural network algorithm model for sinter ore content less than 10 mm is established.The results show that the model prediction effect is very good,it’s accutacy up to0.1237,and the purpose of accurately predicting the content of <10mm grain size of sintered ore is achieved,which provides a method for effectively predicting the content of <10mm grain size of sintered ore.Finally,the XGBoost algorithm is applied to predict the FeO content of sinter and compare it with the prediction effect of the decision tree model.The results show that the XGBoost model has a good prediction effect,and the predicted loss value can reach a minimum of 0.071876,which achieves the purpose of accurately predicting the FeO content and achieves satisfactory results.Figure 36;Table 3;Reference 89...