Research on Fault Early Warning of Wind Turbine Pitch System based on Long Short-Term Memory Neural Network

To reduce the downtime of wind turbine caused by the fault of pitch system that usually has a high failure rate, a fault early warning method based on long short-term memory (LSTM) neural network is proposed. According to the operation characteristics of the pitch system, monitoring parameters recorded by SCADA, such as wind speed, generator speed, rotor speed, wind direction, pitch angle and output power, are utilized to establish the fault early warning model. Compared with the model based on BP neural network, the prediction accuracy of the LSTM model is effectively improved. The weights and thresholds of the LSTM neural network are optimized, and a reasonable threshold is set to determine the fault early warning criterion for the pitch system. To verify the effectiveness of the proposed method, the LSTM model is applied to the fault early warning for a 1.5MW wind turbine pitch system. The results show that the LSTM model can give early warning about two and a half days before the fault occurs, verifying the feasibility of the proposed method.