Research On The Stability Evaluation For Blast Furnace Condition With Data Mining Method

Blast furnace ironmaking is an important part of iron and steel industry.Stable and smooth running of blast furnace can not only realize high production,high quality and low consumption,but also extend the life of blast furnace.In order to achieve the stability and optimization of blast furnace ironmaking process,reduce the energy consumption.It is necessary to evaluate and judge the current condition of blast furnace accurately and timely,then take various appropriate adjustment measures in view of the abnormality.At present,the evaluation of blast furnace mainly based on artificial experience,traditional expert system,neural network,Bayesian network and so on.Elaluation based on the field experience is subjective,and it is difficult to transfer experience.The traditional expert system is based on expert experience to make logical judgments.The amount of knowledge and rules is large,it is difficult to obtain them,and difficult to apply them directly to other blast furnaces.There are supervised data mining models such as neural network and Bayesian network,which can reduce the amount of knowledge and the difficulty of obtaining knowledge,but still not solve the problem of migration.Aiming at these problems,this paper try to use data mining technology to propose an unsupervised furnace condition evaluation method which can evaluate and judge the stability of blast furnace condition based on clustering distance-regression.(1)Based on the analysis of the daily average production data of a large blast furnace in China for 7 consecutive years,it is found that there are some problems in the obtained historical data,such as missing values,outliers,parameter redundancy,parameter inconsistency and so on.Using python language to write relevant programs to complete the data preprocessing processes such as deletion of missing values,identification of abnormal point box graphs,ZScore standardization,data feature conversion,etc.The processed historical data meet the basic requirements of data mining.(2)This paper propose an unsupervised learning method based on cluster analysisdistance-regression for blast furnace condition evaluation.The K-means algorithm is used to cluster the parameters,and the reference vector was obtained by mean value and Z-Score standardization method representing the stability of the blast furnace.The deviation distance between the parameters and the reference vector is calculated by the Euclidean distance.The blast furnace condition index is obtained by fitting the deviation distance with multiple linear regression method.This method can be trained without furnace condition record,which avoids the problems of untrustworthiness of traditional model labels and poor model portability.(3)Referring to the theoretical knowledge of blast furnace smelting and experts'experience,78 state parameters are divided into five attribute sets:blanking,top gas,furnace body heat,air volume and pressure,and hearth.And a two-step cluster-distance regression model is established in this paper(the sub-attribute set is evaluated to obtain the sub-index and then the sub-index is evaluated).Compared with the one-step clustering distance regression model,the two-step clustering-distance-regression model greatly improves the accuracy of furnace condition evaluation.The accuracy of judging abnormal furnace conditions has been improved by 29.63%.(4)The simulation results of the two-step clustering-distance-regression model are analyzed,and the parameter weights are further optimized.Finally,the accuracy rate of the blast furnace condition index for judging the furnace condition is 80.79%,the accuracy rate for judging the normal furnace condition is 86.48%,and the accuracy rate for judging the abnormal furnace condition is 73.06%.