Aiming at the problem that the accuracy rate and false alarm rate are difficult to balance during the maintenance of key components of wind turbines, a wind turbine gearbox fault early warning method based on two-stage multi-dimensional data generation and real-time health index is proposed. To account for the impact of wind speed on the unit′s operational state, a Gaussian Mixture Hidden Markov Model leveraging historical wind speed data is proposed to forecast short-term wind speed. Next, to enhance early warning accuracy, a real-time dynamic threshold-setting method utilizing a two-stage multi-dimensional data generation process is introduced. Based on the predicted wind speed sequence and generator data, the threshold interval for the current oil temperature is established. Finally, the actual gearbox oil temperature values and the output of the health status discriminator are integrated to assess whether the wind turbine gearbox is in an abnormal state. Simulation results using real-world data demonstrate that the proposed method significantly reduces the false alarm rate and provides a potential fault warning for the wind turbine gearbox up to 17 hours in advance.