Fault detection for wind turbines via long short-term memory network

Wind energy is renewable and clean, which plays an important role in energy structure nowadays. However, the operation and maintenance cost of wind turbines (WT) is high, imposing restrictions on its development. Thus, it is necessary to detect early faults in wind turbines. This paper proposes a real-time wind turbine fault detection method. The method utilizes long short-term memory (LSTM) network as the residual generator. The improved network cross-LSTM learns all collected variables to predict the output of wind turbine benchmark, and then the differences between the predicted and true value from the residuals. It also combines the real-time signal processing with LSTM network to classify the faults defined in the benchmark. The simulation results for this method have been compared with other three methods on the benchmark, showing that the former has better accuracy.