Blast Furnace Hot Metal Silicon Content Of Neural Network Prediction Model

Blast Furnace is a great reactor.There are a series of complicated physicochemical reactions in BF.The temperature is the key to ensure the reactions occuring successfully.In normal smelting conditions,the Silicon content in heat metal is direct proportion to the temperature.Therefore,[Si] is an important indicator in BF practice.Forecasting [Si] in heat metal,we can judge whether BF is going on well or not.On the other hand , keeping [Si] in lower level can decrease coke ratio,lower costs of pig iron and bring about profitabilities for smelting absolute steel.At present,the status of BF confronts the powerful challenges from all kinds of iron-smelting ways.For strengthen comptitive abilities of BF,the well-quality crude fuel and stable device must be used.Especially,developing high-efficiency computer controlling system is the most important thing .Smelting procedure is extremely complexity,detected informations are incomplete,response is dull and the range of contrlling is narrow.So developing an IE of BF is a hard work.The way to developing the procedure control model of BF also came through from simple analysis to quantitative calculation, from single method to multi-method combination.With the development of AI,especially the technique of ES and NN have been led in.To uncertainty message ,we can proceed fuzzy deduction and overcome the demerit of BF mathematical model ,which is short of flexibility and adaptability.In particular,NN is used in BF smelting procedure because of its fuction of self-study and fuzzy recognition.Of course,NN has the demerit of"fuzzyness",that is to say,NN cannot explain deductive mechanism itself .Presently,man set about study how to combine the mathematical model,ES and NN to set up procedure control model being suited for production practice of BF.The study wants to develop a Silicon content of heat metal pridiction and control model by use of BP NN and expert rule partly.The functional program of NN predictive model has developed with VB computer language.The correlation datebase is developed with ACCESS.We utilise the date being gained from 5# BF of JinXi Iron and Steel Co.to train the network,to simulate forecasting and control.We also analysed predicting results and analysed the qualititive and quantitative relationship between controlling parameter or medium parameter and silicon content of heat metal.On base of these,we form some control rules and set up a whole silicon content of heat metal prediction and control NN model.The model use fixed method and amendable method to simulate prediction.Under conditions of the same system error,fixed prediction's hit ratio is 75% and that of amendable prediction is 80%,which explain that amendable method is much more better and that the on-line prospect of the model is good.The results show that the model hit ratio is above 90% when the BF is under stable condition.As a result,coke ratio,coke batch weight,blast temperature,blast volume,air permeability index,slag alkalinity,FeO quantity in slag and sulphur quantity in heat metal bring about effects in accord with theoretic analysis of silicon content .