The Development And Application Of Quick Tapping Model

With the growing of steel production and improving of ratio of varieties steel, the controlling of phosphorus, sulfur composition in molten steel was more and more severely. For Wisco third branch, the phenomenon of changing, breaking and returning steel happen for high phosphorus and sulfur at the end of converter process and measures taken not reach the designated position in following process. And measures of desulphurization during tapping were adopted, which cause a waste of cost. So how to predict the content of manganese sulfur and phosphorus in BOF, which provides basis for controlling the following process, has remarkable economic benefits.According to the present situation of computer steelmaking in BOF, the influence factors of c was studied. Using algebra model and the BP neural network algorithm, the prediction model of manganese, phosphorus and sulfur during converter end was established. And fast tapping model was developed with the programming languages of MATLAB, FORTRAN77, and VC++. Therefore, this article mainly draws the following conclusions.(1) Based on the analysis of mechanism in steelmaking process and testing of production data, main input variables, which influence the content of manganese, phosphorus, and sulfur during converter end, were determined.(2) The multiple linear regression models for predicting the content of [Mn], [P], [S] was established respectively after collecting and testing production data. The fitting degree of Multivariate linear regression model was lower. Shooting was not ideal forecast, which couldn't play on the field production instruction significance.(3) The BP neural network model for predicting the content of [Mn], [P], [S] were established respectively. After repeatedly comparative training, its network model structure was fixed. According to the result of test sample predicting by BP neural network model, Fitting degree and shooting of BP neural network model had greatly improved compared with multiple linear regression model.(4) The performance of the established model was studied, the influence of each input variable on BOF steelmaking was analyzed. These all fit the dynamic thermodynamics and metallurgy reaction principle.(5) The [Mn], [P], [S] content of converter end was predicted for field production. The predicted results were that: for prediction of [Mn], the relative error of the prediction in 15% was 91.7%; its absolute error within 0.01% was 93.75%. For prediction of [P], the relative error of the prediction in 15% was 83.75%; its absolute error within 0.002% was 93.75%. For prediction of [S], the relative error of the prediction in 20% was 85.43%; its absolute error within 0.002% was 87.5%.(6) The converter end forecast model is divided into parameter optimization and field prediction two parts. Field prediction part, embedded in the original program, adopted FORTRAN 77 writing. Parameter optimization part, calculated by MATLAB, used VC++ writing. (7) Provisions input parameter scope, standard comparison with prediction result of multiple linear regression models, and properly methods dealing with input process data, were adopted to improve the shooting percent of converter end forecast.