Bf Status Of Intelligent Diagnosis And Prediction Method

The BF (blast furnace) status diagnosis and prediction is importantfor the blast furnace operation. Although expert system has been appliedto BF status diagnosis and prediction very well, disadvantages of difficultexpert acknowledge acquiring and bad transportability limit itsapplication and dissemination. ANNs (artificial neural networks)constructs diagnosis model by learning BF status evidential samples, andsolves the problem of acknowledge aquiring, so it has good behaviour ofapplication and dissemination. But artificial neural netwoks's learningalgorithm is based on Empirical Risk Minimization principle, which isintended for dealing with large sample sizes. In reality, the sample of BFstatus is limited, so ANNs diagnosis model does not work well. In thispaper we present a novel method to solve BF status diagnosis andpredition under a small sample of training instances.To make the diagnosis model perform better under small examples,we present a hierarchy diagnosis model: first, we construct SVM (supportvector machines) classifiers for the initial BF status diagnosis, its outputis the distance to the optimal separating hyperplanes. Second, weconstruct BP artificial neural networks to make further diagnosis.According to the ANNs' output, we make the final diagnosis. The SVMapproach is aimed at solving the classification problems under thecondition of a small sample of training instances, and has bettergeneralization ability. To solve the problem of multi-classification forSVM, we use ANNs to make further diagnosis, it can improve diagnosisaccuracy.BF status can not be represented by a specific detectable parameterexcept furnace overheating, so traditional forecasting approaches aredifficult to use. Because the BF status diagnosis model is to obtain themaximum distance between learning mode data sets, state vectormovement in classification space can reflect changes in the BF status.Therefore, we obtain the historical data's diagnosis results by diagnosismodel, then use this sample data set to train forecasting artificial neural networks, finally, we predict the status of the blast furnace by trainedforecasting ANNs.By simulation, the BF diagnosis and prediction system constructedusing the above method has good results, so it proves that the novelfurnace status diagnosis and prediction method is effective.