Modeling And Prediction Of Blast Furnace Condition Based On Deep Learning

Blast furnace is a closed metallurgical reactor,in which there are many kinds of physical and chemical reaction processes.The direct and indirect measurement of furnace process parameters is the basis of furnace condition analysis.It is of great significance for the stability and smooth operation of blast furnace to realize the modeling and prediction analysis of blast furnace condition.With the rapid development of information technology,the neural network method of furnace condition analysis is becoming more and more important.The traditional neural network is suitable for processing single-mode and local information,and the deep extraction of effective information features of massive data is insufficient.The deep information mining ability of deep learning technology has more advantages than traditional neural network in processing high-dimensional data.Therefore,it is very important to carry out modeling and prediction of blast furnace condition based on deep learning.In this paper,deep learning technology is used to study the parameters of blast furnace ironmaking process,and a logical model describing blast furnace ironmaking process is constructed,and the model is instantiated.The main work is as follows:(1)aiming at the inconsistency and incompleteness of massive process data of blast furnace,principal component analysis(PCA)and factor analysis(FA)are used to process the process data to reduce the main parameters of modeling(2)Combined with the operation,state and index parameters of blast furnace ironmaking process,four logic models describing blast furnace ironmaking process are constructed(3)Combined with deep learning technology,data mining of blast furnace condition is carried out,and convolutional neural networks(CNN)model and long short term memory(LSTM)model are established(4)Aiming at the selection of super parameters in the network,gray wolf algorithm is used to establish super parameter optimization to enhance the information representation ability of the network.Based on the blast furnace process data,the deep learning technology is used to carry out modeling and prediction,and the depth information mining is as follows:(1)the principal component analysis shows that the temperature field data is reduced from 108 dimensions to 20 dimensions,and the principal component of the temperature field can represent the original temperature field information(2)By comparing the four kinds of logic models,it is concluded that the prediction accuracy is improved after PCA processing,and the logic model ? uses artificial modeling of CNN network instantiation to predict the best state parameters of blast furnace,with the average grey correlation degree of 0.857 and the average relative error reduced(3)By optimizing the network structure with gray wolf algorithm,it is concluded that in the modeling benchmark model,the optimal number of all connected layers is 13,the number of nodes is 38,and the average gray correlation degree of state parameters is increased to 0.960.The modeling and prediction of blast furnace condition based on deep learning makes full use of the direct and indirect detection data to realize the model expression of multi-dimensional parameter complex coupling relationship of blast furnace condition,which provides a new method for deep analysis of blast furnace condition.The prediction model of blast furnace condition established in this study has a guiding role for the study and judgment of the trend of blast furnace process parameters.