Driven by the dual carbon objective, the offshore wind power has become a key supporting technology for optimizing the energy structure and promoting green and low-carbon transformation. However, offshore wind turbines operate for extended periods in complex marine environments characterized by high salt spray, high humidity, and strong stochastic loads, resulting in highly variable operating conditions and significant operation and maintenance challenges that urgently require advanced fault diagnosis methods. To address the limitations of existing data-driven fault diagnosis models in capturing deep temporal fault features and in effectively integrating fault mechanism knowledge, this article proposes a large artificial intelligence model-based fault diagnosis method for offshore wind turbines. The clustering and a greedy algorithms are first employed to preprocess the turbine data and perform feature selection, thereby identifying key features closely related to fault evolution and reducing data redundancy and noise interference. Subsequently, a large model is adopted as the backbone architecture and integrated with temporal transformation and classification modules to construct the offshore wind turbine fault diagnosis framework. In particular, to enhance the recognition performance and generalization capability of the proposed model, a prior knowledge loss function incorporating physical constraints and class balancing is designed. Finally, the proposed method is evaluated on the EDP public dataset and a dataset collected from offshore wind turbines in Jiangsu Province. Experimental results show that the proposed approach outperforms typical data-driven models, including convolutional neural networks (CNN) and convolutional neural network Transformer hybrid models (CNN-Transformer), in terms of precision, recall, and F1-Score in fault diagnosis, thereby providing reliable technical support for the safe operation of offshore wind turbines.