Wind power, as an important part of clean energy, plays a key role in the global energy transition. However, wind turbines operate in harsh environments for a long time, and equipment failures occur frequently, which seriously affects power generation efficiency and economic benefits. Aiming at the difficulty of fault identification under complex working conditions of wind turbines, this study proposes a multi-dimensional anthropomorphic condition monitoring method based on CEEMDAN-TCN. The method firstly adopts fully adaptive noise ensemble empirical modal decomposition to decompose the signals of the unit operation data to eliminate the modal aliasing phenomenon, and then utilizes time-domain convolutional network to predictively model the decomposed intrinsic modal components and combines with adaptive crag analysis to realize the fault feature extraction. The experimental results show that the proposed method triggers the alarm 2 h 17 min, 55 min, and 1 h 13 min ahead of time compared with CNN, LSTM, and GRU models, respectively, in gearbox fault warning, and the prediction accuracy is significantly improved. In the pitch system fault diagnosis, the pitch power ratio in the fault state crosses the range of 0.5-2.0, while the normal state is only 1.0-2.0. The method effectively solves the problem of misjudgment and omission of the traditional method through deep mining of spatio-temporal correlation information, and provides a reliable technical support for the intelligent operation and maintenance of wind turbines.
