Abstract: Most of the existing EAF end-point carbon prediction models ignore the influence of the remaining carbon mass in the EAF on the carbon content at the end of the next heat, and the thickness of the slag layer during EAF smelting is thick, and the molten steel splashes violently during smelting. If the composition of molten steel is not uniform, there will be a large deviation between the carbon content detected by the end point sampling of the electric arc furnace and the actual carbon content in the molten steel.If the carbon mass fraction sampled by the electric arc furnace is directly used as the analysis data, the predicted value of the model often deviates from the actual carbon content in molten steel. Aiming at the above problems, the multiple linear regression was used to calculate the yield of iron and steel materials, and the remaining amount of steel in each furnace was calculated. Then, a BP neural network was used to propose a predicition method for the carbon content at the end of the electric arc furnace. Experimental analysis showed that the error between the predicted value of the carbon mass fraction at the end of the model and the actual value was within ±0.02 %, and the hit rate reached 92 %, which had a high prediction accuracy.

Key words: carbon content, multiple linear regression, yield, BP neural network, prediction