Abstract: Based on the reliability operation and maintenance requirements for rotating blades of heavy-duty gas turbines, a lightweight modeling method was developed by reducing the order of numerical simulation results, and combining proper orthogonal decomposition (POD) reduced-order technique and a data-driven machine learning regression fitting approach, so as to realize the rapid prediction of creep life using Larson-Miller method. Results indicate that the proposed reduced-order model can achieve a maximum prediction error of no more than 5% for blade creep life at five critical nodes, while the computational efficiencies of the reduced-order method for temperature field and stress field are 9.00×10⁵ and 1.50×10⁵ times that of traditional simulation method, respectively. Relevant research findings can provide significant theoretical and methodological support for the online operation and maintenance monitoring of rotating blades in heavy-duty gas turbines.