The invention discloses a wind turbine generator blade root load estimation method based on an extreme learning machine. Advantages of information resources of a wind power data acquisition system andrapid learning of an extreme learning machine are fully used to determine output variables according to modeling requirements. Principal component analysis is used to determine system input variables. Through the extreme learning machine, a wind turbine generator blade root load estimation model is established, and a load estimated value in a corresponding input state is calculated through the model. Using an extreme learning machine algorithm to perform a wind turbine generator modeling is simple, and convergence rate is fast, and unit performance and states can be rapidly identified and estimated. The invention provides an effective method for fan modeling research.