Study On Prediction Model Of Properties Based On Artificial Neural Network

With the rapid development of iron and steel industry for a long time,the grade of available iron ore has been declining.In order to alleviate the shortage of raw materials,some resources with high harmful elements and low grade of iron ore and iron-containing solid waste are also used in the sintering process,resulting in a variety of sintering raw materials and unstable composition.As the main charge of blast furnace ironmaking,the fluctuation of raw material composition has a great influence on the production index and the stability of sintered products.Although the sintering pot or micro sintering experiment can grasp the characteristics basis of different ore,and variation of raw material to effect of ore properties,but this methods are tentative method,require a lot of manpower and material resources and time,cannot achieve mastering composition fluctuation influence on sinter properties of fast response.Therefore,it is necessary to establish a prediction model based on sinter related influencing factors,so as to timely feedback the change of sinter quality when the raw material composition fluctuates,which is conducive to stabilizing sintering production and improving sinter quality.In the actual production process,in order to ensure the stable operation of the sintering plant,in a certain period of time,the composition of sintering raw materials fluctuates little,the process parameters are basically unchanged,and the performance of sinter is stable,leading to concentrated data distribution.In addition,the testing process of sintering raw material composition and finished product mineral properties has a large time lag,which leads to the relationship between the input and output of the system can not correspond well.In addition,abnormal working conditions may occur in the field,resulting in large noise in the collected data.Therefore,in this work,the relevant data of sintering experiment was collected in domestic and foreign literatures,which established the sample database of sinter composition and performance index data.In this paper,by using the method of artificial neural network coupled with Fact Sage thermodynamic calculation,the liquid phase and spinel amount generated in sintering process under thermodynamic equilibrium condition were extracted,and the prediction model of sinter tumbler index(TI),yield and low-temperature reduction disintegration index(RDI)based on sintering mechanism was established.Compared with using the production data directly and using the sample database data only using BP neural network to build the prediction model,the neural network model based on sintering mechanism has better data fitting ability and model generalization performance.Firstly,with the contents of Fe O,Al₂O₃,Mg O,Ca O,Si O₂ and Fe₂O₃ as input variables and the mass of the liquid phase and spinel generated in the sintering process under the condition of thermodynamic equilibrium as output variables,the prediction model of phase generation in the sintering process with the neural network structure 6-22-17-2 was established.The mean square errors of the predicted mass of liquid phase and spinel on the model test set are 0.23 and 0.10,respectively,which indicates that the model can accurately predict the phase generated by the theory of sintering process.The mass determination coefficients R² of the output liquid phase and spinel on the model training set were 0.9984 and 0.9993,respectively,which indicated that the predicted value could explain the actual value to a high degree.The determination coefficients on the test set are 0.9983 and 0.9992respectively,which are very little different from the results of the training set,indicating that the model has good generalization ability.The prediction models of sinter TI,yield and RDI with neural network structures of 4-13-8-1,4-5-3-1 and 4-14-1 were established respectively.The mean square errors on the model training set were 6.07,2.27 and 14.23,and the determination coefficients R²were 0.90,0.97 and 0.66,respectively.The mean square errors on the model test set were 19.25,2.44 and 16.95,and the determination coefficients R² were 0.80,0.96 and 0.64,respectively.According to the model,the influence of chemical composition changes on the performance of sinter was also discussed.In the set range,the influence rule of chemical composition changes on sinter strength and RDI obtained by the model was consistent with the general relevant experimental results.The increase of Fe O and Mg O content is not conducive to the formation of liquid phase in sintering process,which leads to the decline of sinter TI,but is beneficial to the suppression of low-temperature reduction degradation phenomenon.The increase of Ca O content is beneficial to the formation of liquid phase in the sintering process and the increase of sinter drum strength.The influence of Ca O content on the low-temperature reduction disintegration index has a peak value.The influence of Al₂O₃ content,Si O₂ content and Fe₂O₃ content on sinter drum index and low temperature reduction pulverization has inflection point,too high will lead to deterioration of sinter performance.The model has a certain guiding effect on sintering test and production.