Abstract: Based on the gas flow and composition of  exhaust gas, which obtained by the exhaust gas analysis, and the carbon integral mathematical model constructed by carbon balance principle, the carbon content in the molten bath could be predicted dynamically. The prediction accuracy of the carbon content in the molten bath dynamically could be improved by the correction on such parameters as the delay time of the exhaust gas information, the volume of exhaust gas, the coefficient of oxygen lance level and the inflection point aand b of  the decarburization  rate  curve. On the basis of dynamic prediction of the carbon content, the dynamic prediction model of the heat temperature could be constructed  on the heat balance theory and the reaction thermodynamic between carbon and oxygen. And the prediction accuracy of the molten bath  temperature could be improved through the correction of decarburization rate inflection point aand b, and different prediction model constructed according to three different stages. The Visual Basic 6.0software and SQL Server 2000  database are employed to develop the  dynamic prediction system about the carbon content and temperature in the molten bath. The data of 46 smelting furnaces were run offline operation by employing this system, and the result of which show that the mass fraction of carbon at the end-point is less than 0.2 %, the  prediction accuracy is 84.8 % as the deviation of the end-point mass fraction of carbon prediction is
less than 0.02 %, and when  the deviation of the end-point temperature is less than 20 ℃, the prediction accuracy is 84.8 %. On the whole, the C-T double hit rate  could reach to 73.9 %, which could basically meet the work-field demand on the end-point prediction.

Key words: exhaust gas analysis, correction at different stages, carbon content, temperature, prediction