Research On The Model For LF Endpoint Prediction

In recent years,with the development of the techniques of clean steel manufacturing and continuous casting,there has been a rapid popularization in the devices and technologies of secondary refining.As a sort of typical equipment for secondary refining,Ladle Furnace(LF)has been widely applied to practical production process because of its excellent properties.At the moment of tapping in LF process,the temperature and compositions of molten steel are required to meet their standards simultaneously,which has therefore made the endpoint control of LF extremely crucial.However,the high cost,long period and low efficiency during the temperature measuring,sampling and testing processes of molten steel have brought great inconvenience to the endpoint control of LF.On this background,the models for predicting the temperature and sulfur content of molten steel in LF are studied in this dissertation in the expectation of compensating the deficiencies caused by traditional measuring methods.In order to research on the methods of molten steel temperature prediction,a mechanism model for temperature prediction is initially established according to the sources,destinations and transfer laws of the heat in LF process.Then based on Genetic Algorithm,the identification of unknown parameters in the model mentioned above is completed.After that,RBF neural network is used for compensating the temperature prediction error generated by the identified mechanism model,and thereby building the hybrid model for the temperature prediction of molten steel in LF.To solve the problem of LF sulfur content prediction,a mechanism model for predicting the steady-state sulfur content is set up through the researches on thermodynamics and kinetics of the desulfurization in LF.Then by taking advantages of the excellent learning and function fitting ability of RBF network,impacts on the endpoint prediction due to the dynamic process of desulfurization are offset,leading to the formation of the hybrid model for the sulfur content prediction of LF molten steel.Finally,factors that affect either the temperature or the sulfur content of molten steel are wholly taken into consideration and a comprehensive model for temperature and sulfur content prediction of molten steel in LF is built to find the potential for improving the efficiency of the prediction further.In this process,a pure data model for comprehensive prediction is established firstly so as to discuss the feasibility of comprehensive prediction.Secondly,the separate mechanism models for temperature and sulfur content prediction are merged with the combination of parameter identification and error learning process.Thus,the final comprehensive hybrid model for LF endpoint prediction is set up,and the prediction result of the model is evaluated via simulation.