Abstract: The planetary gearbox is an important component of the wind turbine transmission system. Its failure rate is high and it is difficult to directly identify the failure situation. In addition, problems such as the difficulty of obtaining fault samples directly and the large environmental noise of the samples increase the difficulty of fault diagnosis. In response to these problems, a gearbox fault diagnosis method based on Bayesian optimization and Wasserstein distance improvement auxiliary classifier generative adversarial network (WAC-GAN) model was proposed. First of all, based on the auxiliary classifier generation adversarial network, a one-dimensional convolutional layer was constructed for the timing characteristics of vibration signals to replace two-dimensional convolutions to improve the efficiency of signal feature extraction; at the same time, batch normalization layers and dropout were added to the generator and discriminator Layer, standardize data structure characteristics. Then, the Bayesian optimization strategy was used to adaptively adjust the discriminator parameters to improve the performance of the discriminator, and the Wasserstein distance was introduced to improve the objective function of the model, and the generator and the discriminator are optimized simultaneously through the game confrontation mechanism, which significantly improves the generalization ability of the model and fault feature extraction capability. Experiments with different fault states of planetary gearboxes under constant speed and variable speed operation are designed. sample sets, this method can achieve sample data enhancement and maintain a good fault recognition accuracy rate, which verifies the advanced nature of the method.