Research On The Development State Model Of Blast Furnace Gas Flow Based On Infrared Image Recognition

Steel has a wide range of uses and is an indispensable material in the construction of infrastructure.The development of the steel industry,as a manifestation of hard power,plays an irreplaceable key role in supporting the development of the national economy and improving the level of national modernization.Through the published data,it can be seen that my country’s steel production has been in a relatively leading position in the world so far.In recent years,the speed of my country’s economic development has been greatly improved,and the output of steel has maintained a trend of continuous growth.After investigation,it was found that blast furnace ironmaking is the most important production method in the iron and steel industry,and its drawbacks have gradually become prominent,such as higher costs,greater pressure on the environment,and more waste.In the past period of underdeveloped science and technology,this production method did help our country to establish a development foundation as soon as possible in a short period of time.However,with the improvement of scientific and technological level and the transformation of my country’s development concept,more emphasis is placed on environmental protection and greenness.Therefore,this traditional method has been suitable for The current development needs urgently need to be changed.Reasonable gas flow distribution is crucial to reducing costs and energy consumption.This paper makes full use of the production data of No.6 blast furnace of a certain steel(one month),and the main research contents are as follows:(1)Feature extraction on the infrared image of gas flow.First,the non-material surface information of the infrared image is removed;secondly,the key features in the image can be extracted by combining the VGG-16 convolutional neural network,and the obtained feature vector is 4096 dimensions;but the dimension is too high to directly identify the transmission needs.The dimensionality reduction process is carried out,and the vector can be reduced to 64 dimensions with the help of the stack auto-encoding network.(2)Cluster the eigenvectors of infrared images.The K-Means algorithm and the Mini Batch K-means algorithm are used to cluster the feature vectors of the infrared images obtained above,and the two clustering algorithms are evaluated according to the average silhouette coefficient.After comparing the average silhouette coefficient data,we know that,The Mini Batch K-means algorithm is obviously inferior to the K-Means algorithm,so it can be concluded that the K-Means algorithm is selected for clustering,and the results are 10 categories.(3)The development status of the gas flow is divided.The K-Means direct division and the K-Means combined division based on singular energy values are used for division,and the Euclidean distance between the cluster center points is used as the evaluation standard,and the K-Means combined division based on singular energy values is obtained.The whole is better than the direct division of K-Means,therefore,K-Means combined division based on singular energy value is selected,and it is divided into 4 categories.(4)Correlation analysis of gas flow development status and gas utilization rate.Two methods are used for correlation analysis.First,the canonical correlation analysis method is used for linear analysis;the second,the mutual information method is used for nonlinear analysis.Through the analysis of the above two methods,it is concluded that the correlation between state 4 and gas utilization rate is relatively high.From this,it can be concluded that adjusting and controlling according to state 4 can significantly improve the utilization effect and avoid waste of resources.Improve intelligence.(5)Predict gas utilization.The neural network models of BP,Elman and Elman based on genetic algorithm are established respectively.In order to obtain the most accurate prediction results,the correlations are also compared,and the mean square error is used as the standard to evaluate the models.Experiments show that the fitting effect of Elman neural network model based on genetic algorithm is better,and it can better meet the experimental goals and achieve ideal results.