A differential privacy-preserving federated learning scheme with predictive maintenance of wind turbines based on deep learning for feature compression and anomaly detection with state assessment

The transport and supply of traditional energy have been greatly affected, which has attracted the attention of countries worldwide and has become an urgent issue. Limited by its own traditional energy shortage and relying on many imported energy raw materials, Taiwan has gradually transformed natural resources into abundant renewable energy through energy transformation, of which wind power has increased up to 1.79 %. It mainly relies on the operation of wind turbines to convert wind power energy into electricity, so the operation of wind turbines is an important issue to maintain stable power collection. This study mainly discusses the wind turbine failure prediction model based on the supervisory control and monitoring system (SCADA) data of 31 wind turbines, and used deep learning and federated learning architecture combined with differential privacy method to find the most suitable model combination through experiments, and establishes wind turbines for state assessment of failure architecture with transmission and prediction, so as to improve wind turbine efficiency and reduce maintenance costs. The proposed method uses 1-dimensional convolutional neural network and bidirectional long short-term memory denoising autoencoder with the transformer (Conv1D-BiLSTM-DAE-TF) for feature compression, and then uses 1-dimensional convolutional neural network (Conv1D), recurrent neural network (RNN), long short term memory (LSTM) and gated recurrent unit (GRU) model for abnormal detection with state assessment. Combined with federated learning and differential privacy architecture for wind turbine failure prediction it can effectively predict performance and privacy security. The results show that it has the best for feature compression convergence performance and fault prediction based on differential privacy for protect model, and construction of models between different wind turbine equipment through the platform.