Study On Mathematical Model Of Sintering Burdening Optimization And Prediction Of FeO Content In Sinter Ore

China’s domestic ore is low grade; impurities and fine-grained, so they must be used with imported ore instead of using separate. The continued high price of imported ore made the domestic steel mills meager profits even losses. So optimizing sintering burdening to using less imported ore has great significances on reducing production costs. In addition, the sources of iron ore in most domestic steel plant are instability, sintering burdening optimization and adjustment of proportion scheme can respond to fluctuations in iron ore supply to ensure stable and sustainable production. Because blast furnace has strict on quality of sinter ore, it is need to predict indicators of sinter and adjust parameters appropriately to ensure qualified indicators. So this paper construct model of sintering burdening optimization, study on sintering burdening optimization in theory to provide reference for industrial production.This paper estimate use value of fine iron ore by evaluation method based on iron-making cost, and substitute use value of fine iron ore for their prize to calculate. And then use genetic algorithm to optimize sintering burdening, the computing result show that this method can solve the optimal proportion scheme which meet the pre-set constraints and achieve the desired goals.The author analyses many factors that affect FeO content in sinter and choose eight technological parameters such as basicity, blend coal ratio, first temperature, pelleting effect, adding water volume, bed depth, ignition temperature and coal gas flow and proportioning of four kinds fine iron ore as the input variables of FeO contenting forecast mode. The author use BP neural network, RBF neural network, and SVM three algorithms to construct forecast model and predict FeO content in sinter ore. The forecast results show that predictive performance of SVM is better than BP neural network, RBF neural network is better than SVM. The average predictive relative error of RBF is2.33%, hit rate which means the predictive relative error below5%is100%. This predictive performance can meet the production requirements. As illustrated, BP neural network is strict to sample quality and greatly affected by initial weights, so it is not available for small sample and multivariable forecast. SVM has the best predictive performance in theory, but its predictive performance depends largely on the selection of parameters and parameter selection based on experiences which lead to local optimal parameters instead of global optimum parameters, so the actual predictive performance is unsatisfied. RBF neural network has best global optimum approach and less parameters which easily to determine the best parameters, so it has best predictive performance.Based on the RBF neural network forecast model, the author firstly discussed single factor relations between basicity, blend coal ratio, adding water volume and FeO content, then draw three-dimensional images graph and contour map. The three-dimensional images can visually show the FeO content variation tendency with the two factors. Contour map can visually determine the value of the variable factors in the different FeO content.