A Study On Coal Injection Quantity Prediction In Blast Furnace Based On The Temperature Trend

Blast furnace ironmaking as one of the three process of iron and steel industry,plays an important role in development and energy saving of the whole iron and steelindustry. Blast furnace coal injection can not only reduce the cost of pig iron, decreaseenergy consumption and reduce carbon emission, but also can improve the hearthworking state to make blast furnace operate stably.Therefore, to build coal injectionforecast model which can provide guidance for blast furnace operator has certaintheoretical significance and application value.The algorithm of support vector machine(SVM) which can not only solve theclassification problem but also can solve the regression problem has become a hot spotof research in recent years. It can solve practical problems of nonlinear, small sample,high dimension, local minimum and etc.In view of the complexity of blast furnace ironmaking process, based on theproduction data of6#blast furnace of Baotou Steel, using data driven modelingtechnology, mainly research on two aspects of furnace temperature trend change andcoal injection quantity prediction of blast furnace. The specific research content of thispaper is as follows:1.By reading a lot of domestic and foreign literature,understand the process of blastfurnace ironmaking, pulverized coal injection in BF and temperature trend change areintroduced, at the same time, the basic theory and development situation of SVM arepresented.2. There are missing value, abnormal value, different scale and other situations inthe large amounts of data acquired at field, it is necessary for data analysis and proces-sing in order to build a more accurate model, mainly including excellent data extraction,data normalization, smooth processing for missing value and abnormal value, andanalysis on the correlation between the parameters.3.There are three types of furnace temperature trend, while the traditional supportvector machine (SVM) is a binary classification model, analyze and compare currentseveral multi-classification methods, choose "OAO" method to establish theclassification model of temperature trend based on SVM, and achieve goodclassification effect.4.Blast furnace pulverized coal injection quantity prediction model are builtrespectively by using BP neural and SVM, compare the two models, results show that the shooting percentage of blast furnace pulverized coal injection quantity predictionmodel established by SVM is higher. Considering the different control strategy underdifferent temperature trend in practical operation, the modeling idea of multiple modelprediction is put forward：use the SVM algorithm to respectively build coal injectionquantity prediction model based on furnace temperature to cool, smooth furnacetemperature, and temperature to the heat, namely, to build coal injection volumeprediction model of multi-SVM based on furnace temperature trend. in the prediction,firstly use classification model of furnace temperature trend to assign the samples to betested to a certain class, and then use the corresponding coal injection quantityprediction model of SVM to predict. By this multi-level prediction algorithm, the modelaccuracy is further improved, the problems of low prediction accuracy and poorgeneralization ability of single model are solved.Analyze and summarize the work done by this paper at the end of paper, and makea prospect for the further research work.