Abstract: To improve energy-saving effect of steam turbine cold end system, a data-driven modeling and operation optimization method for a steam turbine cold end system was proposed. Firstly, steady-state screening was performed on the obtained historical operating data of the turbine. Then, combining mechanism analysis and machine learning algorithms to select features, a power generation prediction model for the steam turbine and a pressure prediction model of its condenser were established. Finally, the operating mode of the cold end equipment was changed and incorporated into the model for optimization. It was applied to a 630 MW unit for actual prediction and model validation. Results show that the established prediction model has good prediction accuracy, can reflect the real operating situation of the turbine in real time, and provide reference for the optimization of the cold end system operation, which is helpful to further promote the energy-saving, emission reduction, and intelligent operation of thermal power units.