Wind turbines face various fault risks during operation, making precise fault diagnosis and prediction crucial for improving wind farm operation efficiency and ensuring system safety. Traditional fault diagnosis methods primarily rely on rule-based models or shallow machine learning algorithms, which often exhibit low accuracy and poor generalization ability when dealing with complex, nonlinear, and strongly time-dependent data. To address these challenges, this paper proposes an encoder-decoder (Seq2Seq) model based on an improved grey wolf optimizer (IGWO) for fault diagnosis and prediction of wind turbines. The model enhances the feature expression of key input moments through an attention mechanism and leverages IGWO to perform global optimization of hyperparameters, improving both prediction accuracy and generalization ability. Compared with traditional models, this approach demonstrates high efficiency and reliability in wind turbine fault prediction, providing technical support for the intelligent operation and maintenance of wind farms.