Abstract: To improve the operating efficiency of steam turbine under low loads, it is necessary to optimize the main steam pressure. A heat rate prediction model was established by support vector machine (SVM) algorithm based on actual operating data of a unit. The improved grey wolf optimization (IGWO) algorithm was used to optimize the hyperparameters of the SVM model. The IGWO algorithm was used to optimize the feasible pressure range under low loads, and the optimized steam turbine sliding pressure curve was obtained and verified by a practical example. Results show that, the heat rate prediction model optimized using the IGWO algorithm can accurately predict the heat rate under low loads. After optimization, the heat rate of the unit is decreased under low loads, especially when the load is 223.83 MW, the heat rate is decreased by 505.96 kJ/(kW·h), presenting the largest reduction. The optimization scheme proposed can effectively improve the thermal economy of steam turbine under low loads.