Abstract: In the production process of hot-rolled strip, the control accuracy of finishing rolling temperature directly affects the microstructure and mechanical properties of the strip, which is a key factor in ensuring dimensional accuracy and good flatness of the strip. Due to the complexity, multivariability and strong coupling characteristics of rolling process, traditional mechanism models have obvious deficiencies in prediction accuracy, which is difficult to meet the requirements for high-precision and high-performance products control.To this end, this article combines a domestic 2 250 mm hot strip mill with the GBDT algorithm and mechanism model to develop a predictive model for the finishing rolling temperature of hot-rolled strips that integrates both mechanistic models and data.This model has both the physical interpretability of a mechanism model and the ability to fully develop GBDT algorithm's advantages in data mining, enabling continuous learning and adjustment of historical data and real-time feedback, thus maintaining the stability and reliability of the model's predictive performance; At the same time, it has self-training and closed loop control functions to achieve automatic closed loop control.After applying the model online, the deviation of long-term genetic prediction for finishing rolling temperature has decreased from 14.55 ℃ to 9.85 ℃. The results show that the model has high calculation accuracy and can meet the requirements for finishing rolling temperature control under different steel grades and working conditions, thereby improving the stability of strip rolling and the precision of head finishing rolling temperature control, enhancing product competitiveness.