Abstract: In order to control liquid steel temperature accurately,forward and backward prediction models for liquid steel temperature in key strategic points of steelmaking process were proposed,based on the analysis of the main influencing factors and the control state of liquid steel temperature in actual steelmaking process. At the same time,to overcome the disadvantages of traditional prediction methods,a hybrid model method based ladle heat status and BP neural network was proposed. The method is based on the ladle heat status tracking model,and gives full consideration to the effects of ladle heat status on molten steel temperature,and combines with BP neural network,which can effectively improve the prediction precision.