Dynamic balance factor-driven fault diagnosis for wind turbine pitch system under data imbalance

As a key executive unit of wind turbine control and protection, the pitch system suffers from poor diagnostic performance and low accuracy due to data imbalance. Therefore, this paper proposes a dynamic balance factor-driven fault diagnosis framework (RelifF-SAO-DBFDformer) for wind turbine pitch system. First, the key fault features of the pitch system are extracted using the ReliefF algorithm to reduce dimensionality and suppress the interference of irrelevant information. Subsequently, a dual-transformer multilevel diagnostic model is used to decouple the diagnostic tasks for healthy state and fault type, and the balance of attention to various categories of samples during model training is achieved by introducing a dynamic balance factor (DBF). Finally, the snow ablation optimisation (SAO) algorithm is used to update the balance factors and optimise the model’s hyperparameters to obtain an efficient fault diagnosis model. In order to verify the diagnostic performance of the proposed method, this paper uses the actual operation data of two wind turbines to carry out several sets of comparison experiments. The results show that compared with the existing methods, the proposed method has significant advantages in diagnostic accuracy and overall performance.