Study On Interval Prediction Method Of Molten Steel Temperature In LF

LF refining process provides the molten steel for the continuous casting at the specified time,whose temperature and composition are all qualified.As one of the most important parameters,the temperature of molten steel can not be continuously measured due to the constraints of measurement technology and cost factors.This problem can be solved by the prediction model.However,the current research on the temperature prediction of molten steel lacks consideration of the substance of the treatment uncertainty.In view of the above problems,this thesis constructs a molten steel temperature interval prediction model,which is compatible with the uncertainty data,and combined with the idea of assemble learning,and has achieved good forecast results.Firstly,starting from the energy balance of the LF refining process,the energy of each part that affects the temperature of the molten steel is analyzed.Based on the simplified energy budget model of each part,a basic prediction model of molten steel temperature according to energy balance is established.However,the entire steelmaking process is a highly complex process.There are many factors that affect the furnace conditions and there are many uncertainties.The basic forecasting model does not consider the uncertainty of the actual data.In order to increase the consideration of the substance of the uncertainty in the forecast work,this thesis analyzes the sources of the uncertainty and proposes corresponding solutions to the sources of the uncertainties in each part.For the uncertainty caused by sampling noise,this thesis establishes an interval prediction model of molten steel temperature in LF refining process based on interval neural network.In the process of establishing the model,the radius disappears and the model over-learns the interval radius.By selecting the appropriate activation function,improving the structure of the loss function and determining its parameters,the above problems are solved.The interval forecasting model can deal with the interval form input and obtain the interval form output which contains the uncertainty information.Finally,owing to the uncertainty caused by the uneven distribution of the sample data,this thesis analyzes the characteristics of the assemble algorithms,selecting the AdaBoost algorithm to construct the assemble prediction model.For the interval regression in this thesis,some adaptive improvements have been done to make it applicable to interval forecasting model.Meanwhile,the convergence of the algorithm is also been analyzed.The simulation results verify the idea of assemble learning algorithm to establish LF molten steel temperature interval forecasting model,which can reduce the generalization error of the overall forecasting model.This not only considers the uncertainty of data distribution,but also effectively reduces the impact of the uncertainty on the model and improves the reliability of the model.