Abstract：Most wind turbine fault early warning methods based on supervisory control and data acquisition (SCADA) data only focus on a specific fault. This paper proposes an early warning method based on improved ensemble empirical mode decomposition (EEMD) and an enhanced vision transformer (ViT) model. The improvements for the EEMD algorithm include dynamic decomposition, component acquisition, and judgment conditions. In terms of dynamic decomposition, the EEMD algorithm is enhanced with the aid of historical operational data from wind turbines. The decomposition window is set up, and the extreme value envelope is calculated based on the local extreme points within the window and the historical decomposition data. Characteristic information can be obtained at different time scales. In terms of component acquisition, the symmetric extension method is employed, utilizing the number of extreme points within the window, combined with piecewise cubic Hermitian interpolation, AR model assistance, sinusoidal fitting, and slope fitting. The final components are obtained by weighted average. In terms of the judgment condition, the standard deviation determines the criterion for obtaining components. The ratio of the window to the length of the reference historical data is added to the convergence criterion, and the number of components is fixed as the decomposition termination condition. The signal is decomposed into a long sequence to obtain more accurate components. The enhanced ViT model comprises a feature matrix splicing layer, a convolutional layer combined with an asymmetric convolution module, a transformer encoder layer with deformable attention, and an output layer consisting of a fully connected network. The model incorporates asymmetric convolution in the convolutional layer, utilizing seven types of convolution kernels. Three of these kernels can fully capture the latent features of different components and the variable characteristics within a single variable of SCADA data, while retaining the time series features. The deformable attention mechanism is added to the transformer encoder layer, which reduces the computational complexity and enhances prediction by focusing on key sampling moments around the reference time. The specific fault warning process of this early warning method is as follows. Firstly, the SCADA data of the wind turbine is cleaned, and the variables are screened. The improved EEMD decomposes each dimensional variable into components, and the feature matrix representing the state of the wind turbine is constructed. Then, the feature matrix is input into the enhanced ViT model for training, and the optimal normal state prediction model is obtained. The model utilizes processing methods specifically designed for the characteristics of SCADA decomposition data in the image tiling and embedding stages. Finally, the residual matrix is obtained by subtracting the predicted feature matrix from the improved EEMD decomposition results. The RMSE is calculated, and an alarm threshold is set to determine whether the wind turbine is operating abnormally, thereby enabling fault warning. SCADA data in the actual operation of wind turbines verifies the proposed method.
[1] Global Wind Energy Council. Global wind report 2023[R]. Brussels: GWEC-global wind energy council, 2024. [2] Qiao Wei, Lu Dingguo.A survey on wind turbine condition monitoring and fault diagnosis: part II: signals and signal processing methods[J]. IEEE Transactions on Industrial Electronics, 2015, 62(10): 6546-6557. [3] 章楷, 胡鹏, 冯江, 等. 基于数据驱动的风电机组故障预警研究[J]. 电工技术, 2024(12): 75-78. Zhang Kai, Hu Peng, Feng Jiang, et al.A data-driven approach to fault prognosis for wind turbines[J]. Electric Engineering, 2024(12): 75-78. [4] 魏乐, 胡晓东, 尹诗. 基于优化XGBoost的风电机组发电机前轴承故障预警[J]. 系统仿真学报, 2021, 33(10): 2335-2343. Wei Le, Hu Xiaodong, Yin Shi.Optimized-XGBoost early warning of wind turbine generator front bearing fault[J]. Journal of System Simulation, 2021, 33(10): 2335-2343. [5] 马良玉, 耿妍竹, 梁书源, 等. 基于Stacking多模型融合的风电机组齿轮箱油池温度异常预警[J]. 中国电机工程学报, 2023, 43(增刊1): 242-251. Ma Liangyu, Geng Yanzhu, Liang Shuyuan, et al.Anomaly warning of wind turbine gearbox oil pool temperature based on Stacking fusion of multiple models[J]. Proceedings of the CSEE, 2023, 43(S1): 242-251. [6] 向玲, 王朋鹤, 李京蓄. 基于CNN-LSTM的风电机组异常状态检测[J]. 振动与冲击, 2021, 40(22): 11-17. Xiang Ling, Wang Penghe, Li Jingxu.Abnormal state detection of wind turbines based on CNN-LSTM[J]. Journal of Vibration and Shock, 2021, 40(22): 11-17. [7] 王依宁, 解大, 王西田, 等. 基于PCA-LSTM模型的风电机网相互作用预测[J]. 中国电机工程学报, 2019, 39(14): 4070-4081. Wang Yining, Xie Da, Wang Xitian, et al.Prediction of interaction between grid and wind farms based on PCA-LSTM model[J]. Proceedings of the CSEE, 2019, 39(14): 4070-4081. [8] 邓飞跃, 郑守禧, 郝如江. 一种轻量化尺度感知调制Swin Transformer模型的轴箱轴承故障诊断方法[J]. 西安交通大学学报, 2024, 58(9): 83-93. Deng Feiyue, Zheng Shouxi, Hao Rujiang.A lightweight scale-aware modulation Swin Transformer network for axlebox bearing fault diagnosis[J]. Journal of Xi'an Jiaotong University, 2024, 58(9): 83-93. [9] 马良玉, 程善珍. 基于支持向量数据描述和XGBoost的风电机组异常工况预警研究[J]. 电工技术学报, 2022, 37(13): 3241-3249. Ma Liangyu, Cheng Shanzhen.Abnormal state early warning of wind turbine generator based on support vector data description and XGBoost[J]. Transactions of China Electrotechnical Society, 2022, 37(13): 3241-3249. [10] 魏书荣, 张鑫, 符杨, 等. 基于GRA-LSTM-Stacking模型的海上双馈风力发电机早期故障预警与诊断[J]. 中国电机工程学报, 2021, 41(7): 2373-2383. Wei Shurong, Zhang Xin, Fu Yang, et al.Early fault warning and diagnosis of offshore wind DFIG based on GRA-LSTM-Stacking model[J]. Proceedings of the CSEE, 2021, 41(7): 2373-2383. [11] 胡爱军, 连俭, 向玲. 基于ACNN和Bi-LSTM的风电机组故障早期识别[J]. 太阳能学报, 2021, 42(12): 143-149. Hu Aijun, Lian Jian, Xiang Ling.Early fault identification of wind turbine based on ACNN and Bi-LSTM[J]. Acta Energiae Solaris Sinica, 2021, 42(12): 143-149. [12] 樊红卫, 马宁阁, 马嘉腾, 等. 基于EMDPWVD时频图像和改进ViT网络的滚动轴承智能故障诊断[J]. 振动与冲击, 2024, 43(11): 246-254. Fan Hongwei, Ma Ningge, Ma Jiateng, et al.Intelligent fault diagnosis of rolling bearing based on EMDPWVD time-frequency images and improved ViT network[J]. Journal of Vibration and Shock, 2024, 43(11): 246-254. [13] 周海成, 石恒初, 曾令森, 等. 基于关系超图增强Transformer的智能站二次设备故障诊断研究[J]. 电力系统保护与控制, 2024, 52(12): 123-132. Zhou Haicheng, Shi Hengchu, Zeng Lingsen, et al.Fault diagnosis of an intelligent substation secondary device based on a relational hypergraph-enhanced Transformer[J]. Power System Protection and Control, 2024, 52(12): 123-132. [14] 陈俊生, 李剑, 陈伟根, 等. 采用滑动窗口及多重加噪比堆栈降噪自编码的风电机组状态异常检测方法[J]. 电工技术学报, 2020, 35(2): 346-358. Chen Junsheng, Li Jian, Chen Weigen, et al.A method for detecting anomaly conditions of wind turbines using stacked denoising autoencoders with sliding window and multiple noise ratios[J]. Transa- ctions of China Electrotechnical Society, 2020, 35(2): 346-358. [15] 刘蔚, 李万铨, 王明峤, 等. 复杂工况下的永磁同步电机典型绕组故障在线诊断[J]. 电工技术学报, 2024, 39(6): 1764-1776. Liu Wei, Li Wanquan, Wang Mingqiao, et al.Online diagnosis of typical winding faults in permanent magnet synchronous motors under complex working conditions[J]. Transactions of China Electrotechnical Society, 2024, 39(6): 1764-1776. [16] 张淼彬, 王丰华, 金玉琪, 等. 基于图像分割及小波脊线的变压器绕组状态检测[J]. 电工技术学报, 2025, 40(2): 640-652. Zhang Miaobin, Wang Fenghua, Jin Yuqi, et al.Transformer winding condition detection based on image segmentation and wavelet ridges[J]. Transa- ctions of China Electrotechnical Society, 2025, 40(2): 640-652. [17] 夏志凌, 胡凯波, 刘心悦, 等. 基于变模态分解的异步电机转子断条故障诊断[J]. 电工技术学报, 2023, 38(8): 2048-2059. Xia Zhiling, Hu Kaibo, Liu Xinyue, et al.Fault diagnosis of rotor broken bar in induction motor based on variable mode decomposition[J]. Transa- ctions of China Electrotechnical Society, 2023, 38(8): 2048-2059. [18] 齐萌, 王国强, 石念峰, 等. 基于时频图与视觉Transformer的滚动轴承智能故障诊断方法[J]. 轴承, 2024(10): 115-123. Qi Meng, Wang Guoqiang, Shi Nianfeng, et al.Intelligent fault diagnosis method for rolling bearings based on time-frequency diagram and vision Trans- former[J]. Bearing, 2024(10): 115-123. [19] 陈尚年, 李录平, 张世海, 等. 基于EEMD-LSTM的汽轮机转子碰磨故障诊断模型及其工程应用[J]. 热能动力工程, 2023, 38(8): 159-168. Chen Shangnian, Li Luping, Zhang Shihai, et al.EEMD-LSTM-based turbine rotor rub-impact fault diagnosis model and its engineering application[J]. Journal of Engineering for Thermal Energy and Power, 2023, 38(8): 159-168. [20] 马永光, 冯勇升. 基于IICEEMDAN-PCA-GRU的风电机组齿轮箱故障预警方法研究[J]. 太阳能学报, 2023, 44(4): 67-73. Ma Yongguang, Feng Yongsheng.Research on fault warning method of wind turbine gearbox based on IICEEMDAN-PCA-GRU[J]. Acta Energiae Solaris Sinica, 2023, 44(4): 67-73. [21] 赵洪山, 林诗雨, 孙承妍, 等. 考虑多时间尺度信息的风力发电机滚动轴承故障预测[J]. 中国电机工程学报, 2024, 44(22): 8908-8920. Zhao Hongshan, Lin Shiyu, Sun Chengyan, et al.Fault prediction of wind turbine rolling bearing considering multi-time scale information[J]. Pro- ceedings of the CSEE, 2024, 44(22): 8908-8920. [22] 张君昌, 赵莉. 一种基于改进EMD的语音去噪方法[J]. 计算机仿真, 2011, 28(8): 397-400, 412. Zhang Junchang, Zhao Li.Speech de-noising method based on improved EMD[J]. Computer Simulation, 2011, 28(8): 397-400, 412. [23] Dosovitskiy A, Beyer L, Kolesnikov A, et al.An image is worth 16x16 words[J]. Transformers for Image Recognition at Scale, 2020, DOI: 10.48550/ arXiv.2010.11929. [24] Xia Zhuofan, Pan Xuran, Song Shiji, et al.Vision transformer with deformable attention[C]//IEEE/CVF Conference on Computer Vision and Pattern Recog- nition, New Orleans, LA, USA, 2022: 1-12.
2025, Vol. 40 
