Research On The Prediction Model Of Parameters In The Process Of LF Refining

The LF refining process is an important link between converter and continuous casting. The mass of adding alloy have a direct influence on the molten steel content of refining end. The power-consumption, as an important energy-consumption index in the process steel production, is one of the indexes that measure the production efficiency of a company. Therefore, it is very important to predict the mass of adding alloy and the amount of power-consumption preciously in the process of LF refining.In the paper, an alloy-adding model and a power-consumption model are established combining the principle modeling and the intelligent modeling. First, data related to the process of LF refining are extracted to establish a new data base from the practical production data in the process of steel-making and continuous casting of Sanming steel plant. The data filtering is executed according to the limiting condition of each parameter in the process of refining to establish a sample data base and a verification data base. Second, a principle model of the mass of adding alloy in the process of refining is established according to the principle of material balance. Combining related data, the prediction of the mass the adding alloy is realized by using the method of adopting the average value of alloy recovery rate of similar furnaces. Third, predicting the reference furnace by solving the similar-rate and near-rate between predicting furnace and sample furnace according to the principle of case-based reasoning. Last, input vectors of the amount of power-consumption predicting model are confirmed by partial correlation analysis. The amount of power-consumption predicting model is trained by using the principle of BP neural network. And the reference furnace is confirmed according to the principle of minimum relative error. Thus, an integrated amount of power-consumption predicting model is established combining the maximum similar-rate and the nearest furnace.The accuracy of the model is validated by200data of verification data base. The result shows:By using the predicting model of the mass of adding alloy, the cases of relative error lower than0.3account for the percentage of86.5. By using the amount of power-consumption predicting model, the cases of relative error lower than0.4account for the percentage of86.5. At the same time, a system platform is designed by using the tool of Visual Basic. The system realizes the functions of importing parameters of the original process of LF refining and exporting the predicting results of the mass of adding alloy, the amount of power-consumption and some other reference parameters.