Abstract: This article proposes a fault data augmentation and diagnosis technique based on grey correlation analysis to address the challenges of small sample size, high noise, and non-stationary data in fault diagnosis of offshore wind turbines. By improving the grey correlation feature extraction method, developing a correlation constrained adversarial generation network, and constructing a multi-level diagnostic architecture, an average diagnostic accuracy of 99.37% was achieved under typical marine conditions, which is 21.52% higher than traditional SMOTE+RF technology.