Abstract
The cost of steelmaking can be saved by $0.1 per ton steel if the temperature of molten steel after RH (Ruhstahl Heraeus) treatment drops by 1 degree, so it is necessary to control the temperature of molten steel after RH treatment accurately in order to satisfy the superheat of molten steel in tundish. Such a temperature control depends on the hit rate greater 90% if the error range is [Formula: see text]. It is more than 10 years since intelligent models were introduced to predict the temperature of molten steel in RH, but such an issue is still far from being solved. Thus, it is necessary to find some methods to solve this problem from the viewpoints of data preprocess, feature selection and data modelling. Numerical results show that there are nine key factors to affect the temperature of molten steel after RH treatment: the temperature, the oxygen mass fraction and the carbon mass fraction of molten steel before RH treatment, the vacuum time, and the oxygen consumption, the oxygen mass fraction after decarburisation, the aluminium consumption, the thickness of the slag layer, the headroom. And these nine key factors can be reduced to six principal components by the principal component analysis method. Especially, data preprocess can deduce the prediction error and improve the hit rate effectively. If the error range is [Formula: see text], the hit rate is up 45% for back-propagation neural network and back-propagation neural network optimised by particle swarm optimisation when the data set jumps from 711 samples to 471 samples, and the hit rate reach 92.2% in the case of back-propagation neural network optimised by particle swarm optimisation after data preprocess and feature selection.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call