Endpoint Prediction Model Of Molten Steel In LF Based On AdaBoost Assembled Learning Algorithm

As an important part of the advanced steel making product’s process in the world, the ladle furnace (LF) steel making is a kind of secondary smelting technique. It aims to add deoxidant and alloy to the molten steel after being smelt in the converter or electric furnace to deoxygenate, desulphurize and uniform the component in order to smelt high quality and special typed steel. The issue on how to control temperature and component of steel on endpoint accurately meets a pressing need. With the help of accurate prediction of LF’s temperature and component of steel on endpoint, it is not only valuable to improve quality of molten steel and reduce costs, but also helpful for operators to choose the most effective control strategy and organize production rationally.This article is based on the research background of LF’s manufacturing technique of100tons of steel in Sanming, Fujian Refining Furnace. Advanced metallurgical model is being introduced here to control the component and temperature more precisely so that people can get the molten steel with the high quality requirements, and it is used to improve both the hit rate of controlling objectives of products and the use of time of refining furnace refractories so as to reduce operating costs.This article respectively establishes prediction models of the temperature and the desulfurization process on the basis of further analysis of the influence factors of temperature and desulfurization theory. And considering the relationships between the two, a mixed prediction model is created. Taking into account for the problem that the BP neural network is easy to fall into local optimal point, the author puts forward a hybrid algorithm with Adaboost to accelerate the process of BP neural network learning and create a hybrid model based on Adaboost-BP to composite Adaboost generalization and BP neural network’s abilities of local search. Practice shows that the algorithm of combination of Adaboost and BP neural network algorithm can avoid local optimum and it has a faster convergence rate. Finally, matlab toolbox is applied to simulate the mixed model, and simulation comparison between Adaboost-BP network and the prediction results of standard BP network is then mentioned.