Abstract: Objective: To couple multi-scale finite element and machine learning models to analyze the damage evolution mechanism of wind power foundation structures. Method: Macro scale continuous medium models and micro scale heterogeneous micro configuration models of wind power infrastructure are established separately. The coupling analysis of macro micro multi-scale models is achieved through interface displacement coordination and stress balance constraints to obtain micro evolution information of stress, strain fields, and damage initiation propagation processes of key parts of the structure under different load conditions; Using the temporal mechanical characteristics obtained from finite element calculations as input variables, a machine learning model based on random forest is constructed. Multi scale simulated sample data is used for model training and hyperparameter optimization, achieving end-to-end nonlinear mapping from local mechanical response to global damage state. Result: Under extreme loads, the damage evolution of key connection areas in wind power foundation structures exhibits significant scale crossing characteristics; The accumulation of local damage can lead to a non-linear decay trend in the overall stiffness and bearing capacity of the foundation, especially after the stiffness degradation threshold exceeds 15%, the rate of decrease in bearing capacity significantly intensifies. Conclusion: The established multi-scale finite element machine learning coupling framework can effectively avoid the computational divergence risk of pure numerical simulation and the lack of physical mechanisms driven by pure data, providing theoretical support for the damage tolerance design and operation decision-making of wind power foundation structures under high-intensity wind and earthquake combined action.