A Data-Driven Approach for Yaw Zero Drift Diagnosis of Wind Turbine

The yaw system is a critical component that ensures the accurate alignment of wind turbines with the wind direction. However, if the wind vane is insufficiently calibrated or aging, the yaw system may experience zero drift in the yaw angle, which results in a decrease in the efficiency of power generation from wind turbines. In this paper, we propose a diagnosis model for yaw zero drift of wind turbines, based on the temporal convolutional neural network combined with a quantile regression algorithm, temporal convolutional network-quantile (TCN-Quantile). Firstly, the raw data are filtered according to the optimal tip-speed ratio to identify the wind speed range suitable for diagnosis. Secondly, to construct wind speed bins, the samples with comparable wind speed densities are clustered using the density-based spatial clustering of applications with noise (DBSCAN) technique. Finally, the TCN-Quantile neural network is used to model yaw angle-active power curves for each wind speed bin to estimate the angle of yaw zero drift, which takes advantage of TCN for serial data processing and the quantile regression algorithm for probability density analysis. The experimental analysis shows that the proposed model has a higher detection accuracy compared with the traditional diagnosis method based on statistics.