Abstract: The end-point temperature of the electric arc furnace is a key production indicator during the smelting process. Taking a 130 t SHARC electric arc furnace as the research object, this study identifies the primary and secondary factors affecting the end-point temperature through correlation analysis based on the principle of energy conservation within the furnace, ultimately selecting ten input variables for the prediction model. To improve prediction accuracy, an L2 regularization term is introduced into the BP neural network, the network structure is optimized, and the effects of different optimizers on model performance are compared. Mean absolute error and hit rate are used as performance evaluation indicators for the model. The results show that with an L2 regularization coefficient of 0.17, the Adam optimizer, and a network structure of 10×14×230×1, the proposed end-point temperature prediction model achieves significant improvement over the unoptimized model. Validation using 90 consecutive sets of actual production data shows that the model achieves hit rates of 88.9% and 95.5% for temperature errors within ±10 ℃ and ±15 ℃, respectively, confirming its practicality and reliability.