Wind turbines operate in harsh environments for a long time, and the gearbox as a core transmission component faces severe reliability challenges. Aiming at the problem of low fault diagnosis accuracy of wind turbine gearbox in high noise environment, this study proposes a fault diagnosis method based on adaptive probabilistic random forest. The method firstly adopts the improved global projection algorithm for feature extraction and dimension reduction of gearbox operation data, which effectively retains the local structural information while taking into account the global features; then introduces the quantum wolf pack optimization algorithm to adaptively optimize the key parameters of the random forest, and constructs an adaptive probabilistic random forest classifier; and finally improves the fault identification capability through the multi-channel data fusion technology. The experimental results based on vibration data of 20 noisy wind turbine gearboxes show that the proposed method performs well in the identification of four states, namely, healthy state, secondary planetary gear ring wear, sun wheel crushing, and primary planetary gear ring wear. The fault identification accuracy after the fusion of both directions reaches 97.17%, which is significantly improved compared with 93.33% in the single X-direction and 95.17% in the Y-direction. Compared with the traditional method, the fault identification rate of this method reaches 92%, which is significantly better than the 84% of the support vector machine and the 89% of the traditional random forest, proving the effectiveness and superiority of the proposed method in the fault diagnosis of gearboxes of high-noise wind turbines.