Abstract: Aiming at the degradation fault characteristics of wind turbine gearbox, an intelligent diagnosis and predictive maintenance method integrating multi-source data is proposed. Construct a CNN-LSTM classification model and a GRU-RUL prediction network, establish a health index evaluation system, and introduce a scheduling optimization mechanism based on Bayesian risk. Through engineering empirical verification, the proposed scheme has engineering adaptability and deployment value in terms of fault recognition accuracy and maintenance decision.