Research On Dynamic Characterization Model Of Blast Furnace Profile Management By Data-driven

Blast furnace profile management is a comprehensive output form which reflects the input parameters of blast furnace,such as the characteristics of blast furnace burden,the distribution of air flow,the cooling system,the state of slag skin,the structure of cohesive zone and so on.It is of decisive significance to maintain reasonable blast furnace profile management for high quality,high yield,low consumption and furnace longevity.Firstly,this paper improved the traditional Pauta criterion by adding local time window,and constructed the global+local Pauta criterion to screen outliers in the data set.On the basis of considering the advantages and disadvantages of clustering algorithm and the process characteristics of the blast furnace production,the K-Means clustering algorithm and Two Step clustering algorithm were selected,and the clustering evaluation index was introduced to compare the application effect of K-Means and TwoStep clustering algorithm.The results of clustering evaluation showed that,on the basis of the data set taken in this paper,K-Means clustering algorithm is obviously superior to TwoStep clustering algorithm,and the optimal clustering result is determined when the number of cluster is six.Based on the optimal clustering results,a new data set was imported,and the threshold values of various blast furnace profile managements were set according to the distance between the points within the class and the clustering center.The distance between each data point and different clustering centers was calculated,and the data was distributed.A new class was established for the data points exceeding the threshold,and the dynamic update of blast furnace profile management was realized.Secondly,aiming at the problem that it is difficult to quantify some factors affecting blast furnace profile management,the thickness of slag skin was calculated by onedimensional steady-state heat transfer method,and the heat flux intensity was determined by the flow rate of cooling water and the temperature difference of cooling wall water.Based on the softening and melting experiment,the data variation characteristics of the cohesive zone were found.From the perspective of data drive,the process characteristics of the cohesive zone were integrated,and the random flow behavior of the gas in the cohesive zone roots position was captured by using the temperature change rate of the cooling wall thermocouple over time,so as to determine the cohesive zone roots position.Moreover,according to the change characteristics of cooling wall temperature when the slag skin falls off,a rule to judge the slag skin falls off was established.The slag skin fall-off index and marginal airflow development index were further extracted.Finally,the dynamic variation rule of blast furnace profile management was summarized.In order to represent the changes of blast furnace profile management,the sparse principal component analysis was used to complete the establishment of the evaluation model of blast furnace profile management,and then blast furnace profile management was processed by Python programming language to determine the dynamic change relationship of different types.On the basis of the dynamic variation relationship of blast furnace profile management,combined with the characteristics of cooling wall temperature distribution,slag skin,the cohesive zone roots position,air flow distribution to explore the factors affecting blast furnace profile management.On the basis of determining the dynamic change state law of blast furnace profile management,the directed graph in the process of dynamic change of blast furnace profile management was established,and the edge of the directed graph was endowed with physical meaning,and the weighted directed graph adjacency matrix was constructed.It included the adjacency matrix of blast furnace profile management change,slag peeling index adjacency matrix,edge air flow development index adjacency matrix,cohesive zone roots position adjacency matrix and first,second and third principal component adjacency matrix.According to different path requirements,the path planning of the shortest adjustment time path,the furnace condition stability path and the optimal path of technical evaluation index were completed,which provided guidance for adjusting and operating blast furnace profile management in the blast furnace ironmaking process.In this paper,starting from blast furnace profile management,data-driven as the main means,on the basis of integrating the ironmaking process of the blast furnace,the characterization methods,influencing factors,the law of change and other aspects were discussed.The study of blast furnace profile management was extended from the traditional static model to the dynamic process,which is of great significance for maintaining and adjusting the reasonable blast furnace profile management and ensuring the blast furnace ironmaking process.