The Prediction Model And Application Of Tang Shan Stainless Converter Smelting Terminal

In Tangshan stainless stock limited company, the endpoint control of 100t converter smelting adopts the traditional way, which is judged by steelmaking workers through the flame and their experience, but this way is blind, because the hit rate at the end is low, more winds will be needed, and the smelting circuit is long. If we can establish an accurate forecast model of the converter end point, then controlling the production of smelting by computers will become a truth, then productivity will be raised, production costs will be reduced, and the product quality will be improved.Aiming at the production condition of Tangshan stainless company, this essay studied the factors that affect the temperature of converter smelting end point and the content of P and S, confirmed the control variables of the forecasting model of the converter smelting end point. As a tool, Visual Basic programmed language is used, the BP algorithm is improved, the three-tier BP neural network forecast model of the converter smelting end T, P, C and S based on neural network is established. As samples, 300 furnaces of stainless steel production data are chosen at the scene of a factory in Tangshan and the prediction results of the model close to and reach forecast accuracy of the dynamic control model through training the model. The 100 Furnaces of data can be stored as samples for study. Training and making them update make the model adapt to the changes of production conditions and improve the adaptability of the model.Models have been done trials on converters of 100t at Stainless Steel in Tangshan, and the results are very good. When the network structure of the forecast model of NN enacts that the network layer is three and input parameter is 27, there are 30 nodes in the containing layer, study speed is 0.6, and momentum gene is 0.1. The forecast result of the network to the end-point temperature and the component content will be the best. For the objective error of±0.005% hit rate about learning samples of phosphorus and sulfur, end-point forecast model reached more than 85 percent, for the prediction error of±0.015% hit rate of the carbon content, it is 80 percent and for the temperature forecast within±10℃, the hit rate reached 90 percent.