| The end-point control is an important operation in the later period of LDconverter steelmaking.Artificial experience control method is still used in the TTMGsteel plant, the end-point hitting rate is very low.As a result,it prolongs the smeltingtime,increases the metal consume and influences the steel quality.At present, theadvanced control method is the combination of static and dynamic control.Currently the mechanism model, statistical model and artificial neural networkmodel are in common use. In practical applications, the mechanism model is involvedin many parameters that are difficult to measure and control, so the accurate modelingis still very difficult. If the converter smelting characteristic has the goodreproducibility and the production condition and the craft operation method are stable,we can use the statistical model to predicte molten steel final temperature and carboncontent. Because the converter steelmaking process is highly complex and random,and ANN has the very strong self-learning ,fault to learn and inference capability, itprovides a new implement way to the static control of the LD steelmaking process.Based on the 30t production data in converter steelmaking process with theTTMG, firstly the blowing oxygen and ore quantities models of final static control inconverter steelmaking process were established and the prediction accuracy was0.58% and 10.4% respectively. Due to the complications effect on the finaltemperature and carbon content of molten steel, we used the artificial neural networkmodel with the Levenberg-Marquardt algorithm, the 257 production data were used tomodel training and the other 100 were used to predict the final temperature andcarbon content, the endpoint hitting ratio proposed model was 55% under steeltemperature is from 1646℃ to 1698℃ and the carbon content from 0.033% to0.128%.The application results show that the precision of predicting the finaltemperature and carbon content has been increased remarkably. |
PDF Full Text Request