Study On Prediction Modeling Of The Sinter Chemical Component Based On Deep Learning

Sintering is one of the important processes of iron and steel smelting.Sinter is the main raw material of blast furnace ironmaking in China,and the stability of sinter chemical composition directly influences the blast furnace condition and the quality of iron and steel products.The sintering process is a dynamic system with complex characteristics of large lag,strong coupling and strong nonlinearity,which makes it difficult to control the stability of chemical composition of the sinter.Therefore,it is of great significance to realize accurate prediction for sinter chemical compositions.At present,the traditional shallow network algorithm has not fully reveal the essential law of the sintering process,and the precision of the prediction is not high enough to meet the actual production demand.And in the actual production process,the use of artificial assay has led to a serious lag in data detection,and the chemical composition of sinter can not be predicted in real time.In order to solve the above problems,a prediction model for sinter chemical composition based on Deep Belief Network(DBN)is proposed in this paper,and the on-line prediction system is realized by using prompt gamma-ray neutron activation analysis(PGNAA)technology to realize the real-time prediction of the chemical composition of sinter.Firstly,through in-depth analysis of the sintering process,the model prediction parameters are determined in this paper,and the prediction algorithm is analyzed and the chemical composition prediction model of sinter based on DBN,the selected algorithm,is established.Secondly,grey relational analysis is used to determine the input parameters of the model,and removing the abnormal data of sample and normalization process.Based on this,design the structure of the DBN model and build a prediction model of sinter chemical composition based on DBN,simulation analyzing for model based on actual production data,and compared with the simulation result of BP neural network model and SVM model and other common shallow models.The results show that the DBN model proposed in this paper has high prediction accuracy and has obvious advantages compared with other methods.Finally,aiming at the problem of large delay in off-line detection of sintering process,PGNAA technology is used to realize the on-line detection of components,providing realtime data for DBN prediction model,and the on-line prediction system is developed.The simulation run result of system shows that the system can accurately predict the chemical composition of sinter,and provides a new idea for the stable control of sinter chemical composition.