Abstract: Aiming at the characteristics of large inertia, large time delay, time change and complex working conditions of the main steam temperature object of the supercritical unit, we proposed a main steam temperature predictive control strategy based on the digital twin model. We used the control strategy to predict the main steam temperature value through the digital twin model of main steam temperature, used feedback correction to correct the predicted value, and used the golden section method to roll the optimization objective function to output the optimal control amount. The established main steam temperature digital twin model, including the main steam temperature system transfer function model and the Xgboost-LSTM deviation model, used the transfer learning theory to achieve online update, with high prediction accuracy, efficient update ability and strong robustness. Taking the final stage superheating system of a 1000MW supercritical unit as an example, we conducted a simulation experiment. The results show that the main steam temperature predictive control strategy based on the digital twin model effectively improves the control quality and robustness of the main steam temperature system, and has good load adaptability.