Study And Application Of Prediction System Of The Sinter Quality Based On Combination Model

Sintering is essential in the procedure of iron and steel smelting process.Accounting for more than 70% of the raw material for blast furnace,the quality of sinter has a direct influence on the state of blast furnace and quality of final products.The sintering process is a dynamic system featuring large lag,strong coupling and strong nonlinearity.So,it is hard to guide to iron ore proportion in sintering according to the quality inspection results.Therefore,the establishment of the sinter quality prediction model will be useful to forecast the quality of sinter in advance,pointedly adjust sintering process parameters and reduce the sinter quality fluctuation.In sintering,artificial test often leads to the large lag of data detection,for which the sinter quality can not be accurately real-time predicted.On the basis of the mechanism and characteristics of the sintering process,the main influence factors of the quality of sinter are given and an integrated prediction model based on grey system theory and BP neural network is put forward in the paper.The main study results are as follows.Firstly,grey correlation analysis is applied to sintering process parameters and the sinter quality parameters,and therefore the key influence parameters of the quality of sinter are decided.Later,Limit Breadth Average Filter is applied to dealing with the detection data of those parameters.Second,the grey GM(1,1)prediction model is established on the basis of the detection data of sinter quality parameters.And the BP neural networks model is created based on the detection data of key influence factors and sinter quality parameters.Finally,the weighting coefficient of combination model is calculated using the information entropy,the minimum forecasting error sum of squares as the optimization goal and non-optimization goal methods.By weighting combination,sinter quality combination predict models are respectively established to forecast accurate values of TFe content,FeO content,basicity and tumbler index.Besides,five assessment indexes,such as error sum of squares,mean square error,mean absolute error,mean absolute percentage error and precision,are also applied to estimating the forecasting results of three kinds of combination predict model and two kinds of single predict model.The results show that the combination predict model can improve the accuracy of forecasting results,and combination forecast model with the minimumforecasting error sum of squares as the optimization goal has significant advantages comparing with other prediction methods.Sinter quality prediction system is developed based on the sintering process in a domestic iron and steel enterprise.Users can invoke appropriate prediction model according to the actual production situation.Model prediction accuracy is high.The running results show that: the system can exactly predict sinter quality and guide the optimization control of the sintering process.