In order to improve the control effect of reheat steam temperature (RST) during the frequent load change of coalfired units,a RST predictive optimization control approach based on machine learning was proposed. Firstly,the RST prediction model was developed with the historical variableload operation data by using the eXtreme Gradient Boosting (XGBoost) algorithm,and the model parameters were optimized with the random search method to improve its prediction accuracy. Based on the final welltrained model,an improved grey wolf optimizer (IGWO) was employed to realize predictive optimization control of RST by searching the realtime optimal instructions of the fluegasside reheat baffle opening and the steamside waterspray desuperheating valve. Optimization control simulation tests were carried out with a fullscope simulator. The experimental results show that the intelligent predictive optimization control scheme proposed in this paper can effectively improve the control effect of RST,and significantly reduce the amount of desuperheating water spray,which helps to improve the economy of the unit.