Temperature Prediction Of Ladle Furnace Based On ADABOOST Algorithm And ELM Algorithm

With the continuous development of industry,the precision of steel requirements improves.Ladle furnace just meet the requirements,LF refining greatly promote the new kinds of steels refining and new technology research and development.LF furnace maintained reduction atmosphere,argon stirring,submerged heating and synthetic slag refining unique function.The structure of LF is quite simple,with a variety of effective furnace refining method,which can also improve the purity of molten steel and meet requirements of the continuous casting of steel composition and temperature of the liquid.For the Ladle refining process,to predict the temperature of the ladle refining furnace is quite important.However,the impact of factors of temperature is complex.How to accurately predict the furnace temperature is the problem.First of all,according to the LF furnace smelting process and its characteristics,from the angle of total energy balance in the furnace of the fluctuation of LF furnace smelting process energy analysis,I determined the factors that affected the furnace temperature.Ladle refining furnace temperature prediction model for modeling:the influence of temperature variation of factors are inputs to the model,the predicted temperature are the output of the model.Modeling methods in this subject are Extreme Learning Machine(ELM)and Partial Least Squares(PLS)method combined to be the establishment of the basic temperature prediction model.PLS method eliminates data correlation and ELM can be calculated quickly.Aiming at the problems of shortage of the forecasting accuracy of the single model,I proposed the ELM-PLS algorithm which based on AdaBoost.This model has a significant reduce in predict error.Finally,according to the uncertainty of the weight threshold model,I use the genetic algorithm(GA)to optimize the AdaBoost ensembled ELM-PLS model.Respectively,I use the genetic algorithm to optimized extreme learning machine’s initial weights and also to optimize AdaBoost threshold,instead of the extreme learning machine’s random factors and threshold numerical experience.This model results in improving the stability of the overall forecast model,and promoting the accuracy of the prediction.