Abstract: Billet heating is a key process in hot rolling production, and its quality directly affects the performance of final products and production energy efficiency. In this study, a three-dimensional transient temperature field model of billets in a walking beam reheating furnace was first established by using ANSYS finite element software, which revealed the law of non-uniform temperature rise inside billets and the influence of endothermic phase transformation on the temperature rise rate. Secondly, based on the finite element simulation data, samples were generated by the Box-Behnken Design experimental method, and a BP neural network prediction model was constructed. Finally, the response surface methodology was applied to optimize the heating process parameters. Taking the billet exit surface temperature（T_d≥1 000 ℃）, the temperature difference between core and surface（T_z≤50 ℃） and energy conservation and consumption reduction as the objectives, the optimal combination of process parameters was determined as follows: preheating section at 900 ℃, heating section Ⅰ at 950 ℃, heating section Ⅱ at 1 350 ℃, heating section Ⅲ at 1 350 ℃, soaking section at 1 095 ℃, and billet running speed at 0.006 m/s. Under these conditions, the predicted exit surface temperature of the billet is 1 002 ℃ and the temperature difference between core and surface is 49.36 ℃, both of which meet the process requirements. The results show that this method can effectively simulate and predict the billet heating process, which is of great significance for the intelligent control of the heating process. This study innovatively integrates finite element numerical simulation, BP neural network prediction and response surface methodology optimization to construct a closed-loop system of "mechanism-data-optimization", accurately considers the endothermic effect of solid-state phase transformation of billets, and breaks through the limitations of traditional models.