Abstract:

The stator current and voltage fluctuation caused by early inter-turn short circuit of permanent magnet wind turbine is weak, which makes fault identification difficult and prone to miscalculation. To solve this problem, the convolutional neural network and bidirectional long short-term memory network (CNN-BiLSTM) fault diagnosis method based on attention mechanism is proposed. The zero sequence components of generator current and voltage are taken as fault features, and the CNN and BiLSTM fault diagnosis models are used for feature self extraction. Attention mechanism is introduced to evaluate the weight of features at different time, and the weight of key features in fault diagnosis is strengthened based on this. The results of numerical examples show that the proposed method can realize the early inter-turn short circuit fault diagnosis of permanent magnet wind turbine generator, significantly improve the identification accuracy compared with traditional methods, and reduce the time spent in model training.

Key words: wind power generator, inter-turn short circuit fault, fault diagnosis, deep learning, attention mechanism