Research And Application Of Ladle Furnace Temperature Prediction Model And Alloy Charging Model

Ladle furnace(LF)refining is the main means of off furnace refining.Its main task is to accurately control and adjust the temperature and composition of molten steel.Therefore,it is necessary to obtain the temperature and composition information of molten steel in a timely manner.At present,most steel plants at home and abroad use operators to manually measure the temperature and sample to obtain information and calculate the feeding amount.Due to the limitation of temperature measurement and sampling elements,the information cannot be updated in time.At the same time,multiple temperature measurement and sampling operations not only cause waste of resources,but also have great danger.Therefore,in order to realize "one key steelmaking",it is necessary to timely estimate the real-time temperature and composition information of molten steel and establish an accurate alloy feeding model.Aiming at these problems,this thesis studies the production mode of 120 t LF in a steel plant,analyzes the factors affecting temperature and composition,and establishes temperature prediction model and alloy charging model.Based on these two models,LF temperature prediction module and alloy feeding module are designed and embedded into LF refining system to help enterprises optimize refining process.The main work of this thesis is as follows:(1)In view of the problem that the temperature cannot be obtained in real time,this thesis studies the existing production mode of the refining furnace,analyzes the factors that affect the temperature and composition,and preprocesses the collected data.Based on the poor fitting ability of traditional neural networks on small sample sets,this thesis selects generalized regression neural networks to establish a model.This model can improve the real-time prediction and ensure the accuracy.Compared with BP networks,RBF networks and support vector regression networks,it proves that the prediction performance of generalized regression networks is good and the accuracy is high.(2)In view of the alloy composition of molten steel,the extreme learning machine is used to predict the element yield in the refining process of molten steel.To solve the problem of unstable model performance caused by random selection of weights and thresholds in the extreme learning machine itself,genetic algorithm is used for optimization.Compared with BP model and ELM model,GA-ELM model has better stability and prediction accuracy.At present,when researchers study the prediction of alloy yield,the initial temperature is used instead of the real-time temperature because it is impossible to obtain the real-time temperature.In this thesis,the correlation between temperature and prediction results is studied,and a GRNN-GA-ELM element yield prediction model is proposed.The model can predict the temperature at the time of feeding through the temperature prediction model to solve the problem that the realtime temperature cannot be obtained.The experimental results show that the stability and prediction accuracy of the GRNN-GA-ELM model proposed in this thesis are improved.(3)In view of the problem of operators’ manual calculation of feeding in the smelting process,this thesis improves the current single linear programming model based on the element yield prediction model to make it closer to the actual use of steel plants,and establishes the alloy feeding model based on the multi-objective programming method.The model can calculate the optimal feeding formula under the constraint of target composition.This thesis completes the application of the model through Lab VIEW platform and python programming language.