In the prediction of end-point molten steel temperature of LF, the influence of some factors is nonlinear. The prediction accuracy will be affected by directly inputting these nonlinear factors into the data-driven model. To solve this problem, an improved case-based reasoning model (CBR_HTC) was established through the nonlinear processing of these factors by calculating the heat transfer of the ladle with software Ansys. The results show that CBR_HTC model improves the prediction accuracy of end-point molten steel temperature by 5.33% and 7.00% compared to original CBR model, and 6.66% and 5.33% compared to BPNN model in the range of [-3,3] and [-7,7]. The MAE and RMSE values of CBR_HTC model are also lower. It is verified that the prediction accuracy of the data-driven model can be improved by coupling the mechanism model with the data-driven model.