基于循环DBSMOTE和多头注意力机制的风力发电机故障预警体系

綦方中; 蔡阮昊; 曹柬

doi:10.19912/j.0254-0096.tynxb.2025-0109

PDF(2520 KB)

太阳能学报 ›› 2025, Vol. 46 ›› Issue (11) : 772-782. DOI: 10.19912/j.0254-0096.tynxb.2025-0109

基于循环DBSMOTE和多头注意力机制的风力发电机故障预警体系

綦方中^1,2, 蔡阮昊¹, 曹柬¹

作者信息 +

WIND TURBINE FAULT WARNING SYSTEM BASED ON CYCLIC DBSMOTE AND MULTI-HEAD SELF-ATTENTION MECHANISM

Qi Fangzhong^1,2, Cai Ruanhao¹, Cao Jian¹

Author information +

文章历史 +

摘要

在结合传统卷积神经网络（CNN）和长短期记忆网络（LSTM）优势的基础上,提出一种基于循环DBSMOTE和多头注意力机制的风力发电机故障预警组合模型与应用体系。针对风力发电机运行数据集不平衡比率较高的问题,提出并使用一种循环DBSMOTE算法对数据集进行平衡处理。在组合预警模型中,使用CNN和LSTM算法来提取平衡处理后输入数据的空间维度特征和时间维度特征以增加数据挖掘的深度,同时引入多头注意力机制,极大地提升了模型的训练精度和预警性能。通过对采集自实际风场的风力发电机运行数据进行训练,模型在实验中实现了提前期长达32 h的故障预警能力。进一步地,通过消融实验、对比实验和十折交叉验证实验,表明所提出的组合预警模型具有较高的有效性和稳定性。

Abstract

This study proposes a hybrid fault early-warning model and architecture for wind turbines, integrating the advantages of traditional convolutional neural networks （CNN） and long short-term memory networks （LSTM）, coupled with a cyclic DBSMOTE algorithm and a multi-head attention mechanism. First, a cyclic DBSMOTE algorithm is developed to balance highly imbalanced wind turbine operational datasets, mitigating the bias caused by extreme class distribution. In the hybrid early-warning model, CNN and LSTM are employed to extract spatial and temporal features from the balanced data, respectively, enhancing the depth of data mining. Additionally, a multi-head attention mechanism is introduced to prioritize critical temporal-spatial correlations, significantly improving the model’s training accuracy and early-warning performance. Through hyperparameter optimization, the model achieves fault early-warning capabilities with a lead time of up to 32 hours in experimental validation. Further validation via ablation studies, comparative experiments, and ten-fold cross-validation demonstrates that the proposed hybrid model exhibits superior effectiveness and stability in wind turbine fault early warning.

导出引用

綦方中, 蔡阮昊, 曹柬. 基于循环DBSMOTE和多头注意力机制的风力发电机故障预警体系[J]. 太阳能学报. 2025, 46(11): 772-782 https://doi.org/10.19912/j.0254-0096.tynxb.2025-0109

Qi Fangzhong, Cai Ruanhao, Cao Jian. WIND TURBINE FAULT WARNING SYSTEM BASED ON CYCLIC DBSMOTE AND MULTI-HEAD SELF-ATTENTION MECHANISM[J]. Acta Energiae Solaris Sinica. 2025, 46(11): 772-782 https://doi.org/10.19912/j.0254-0096.tynxb.2025-0109

中图分类号： TP183

参考文献

[1] HUTCHINSON M, ZHAO F. Global wind report2023[R]. GWR23, 2023.
[2] 谷晓娇, 陈长征. 基于VMD和QPSO-SR的风力发电机轴承故障提取方法[J]. 太阳能学报, 2019, 40(10): 2946-2952.
GU X J, CHEN C Z.Fault extraction method of wind turbine bearing based on VMD and QPSO-SR[J]. Acta energiae solaris sinica, 2019, 40(10): 2946-2952.
[3] LIAN J J, CAI O, DONG X F, et al.Health monitoring and safety evaluation of the offshore wind turbine structure: a review and discussion of future development[J]. Sustainability, 2019, 11(2): 494.
[4] LECUN Y, BENGIO Y, HINTON G.Deep learning[J]. Nature, 2015, 521(7553): 436-444.
[5] CASTELLANI F, ASTOLFI D, NATILI F.SCADA data analysis methods for diagnosis of electrical faults to wind turbine generators[J]. Applied sciences, 2021, 11(8): 3307.
[6] 丁显, 韩宁宁, 滕伟. 基于深度学习的直驱风力发电机故障辨识[J]. 可再生能源, 2018, 36(10): 1535-1540.
DING X, HAN N N, TENG W.Fault identification of direct drive wind turbine based on deep learning[J]. Renewable energy resources, 2018, 36(10): 1535-1540.
[7] 刘家瑞, 杨国田, 杨锡运. 基于深度卷积自编码器的风力发电机故障预警方法研究[J]. 太阳能学报, 2022, 43(11): 215-223.
LIU J R, YANG G T, YANG X Y.Research on wind turbine fault warning method based on deep convolution auto-encoder[J]. Acta energiae solaris sinica, 2022, 43(11): 215-223.
[8] 郭俊锋, 王淼生, 孙磊, 等. 基于生成对抗网络的滚动轴承不平衡数据集故障诊断新方法[J]. 计算机集成制造系统, 2022, 28(9): 2825-2835.
GUO J F, WANG M S, SUN L, et al.New method of fault diagnosis for rolling bearing imbalance data set based on generative adversarial network[J]. Computer integrated manufacturing systems, 2022, 28(9): 2825-2835.
[9] 李慧芳, 徐光浩, 黄双喜. 基于主动生成式过采样和深度堆叠网络的轴承故障诊断[J]. 计算机集成制造系统, 2023, 29(1): 146-159.
LI H F, XU G H, HUANG S X.Active generative oversampling and deep stacking network based bearing fault diagnosis approach[J]. Computer integrated manufacturing systems, 2023, 29(1): 146-159.
[10] BUNKHUMPORNPAT C, SINAPIROMSARAN K, LURSINSAP C.DBSMOTE: density-based synthetic minority over-sampling TEchnique[J]. Applied intelligence, 2012, 36(3): 664-684.
[11] 王朋飞, 刘长良, 徐健, 等. 基于图注意力和时间卷积网络的风电齿轮箱故障预警方法[J]. 电子测量与仪器学报, 2023, 37(8): 204-213.
WANG P F, LIU C L, XU J, et al.Wind turbine gearbox fault warning method based on graph attention and temporal convolutional network[J]. Journal of electronic measurement and instrumentation, 2023, 37(8): 204-213.
[12] 黄荣舟, 汤宝平, 杨燕妮, 等. 基于长短时记忆网络融合SCADA数据的风电齿轮箱状态监测[J]. 太阳能学报, 2021, 42(1): 235-239.
HUANG R Z, TANG B P, YANG Y N, et al.Condition monitoring of wind turbine gearbox based on LSTM neural network fusing SCADA data[J]. Acta energiae solaris sinica, 2021, 42(1): 235-239.
[13] CHEN H S, LIU H, CHU X N, et al.Anomaly detection and critical SCADA parameters identification for wind turbines based on LSTM-AE neural network[J]. Renewable energy, 2021, 172: 829-840.
[14] 赵小强, 张亚洲. 利用改进卷积神经网络的滚动轴承变工况故障诊断方法[J]. 西安交通大学学报, 2021, 55(12): 108-118.
ZHAO X Q, ZHANG Y Z.Improved CNN-based fault diagnosis method for rolling bearings under variable working conditions[J]. Journal of Xi’an Jiaotong University, 2021, 55(12): 108-118.
[15] 向玲, 王朋鹤, 李京蓄. 基于CNN-LSTM的风力发电机异常状态检测[J]. 振动与冲击, 2021, 40(22): 11-17.
XIANG L, WANG P H, LI J X.Abnormal state detection of wind turbines based on CNN-LSTM[J]. Journal of vibration and shock, 2021, 40(22): 11-17.
[16] KONG Z Q, TANG B P, DENG L, et al.Condition monitoring of wind turbines based on spatio-temporal fusion of SCADA data by convolutional neural networks and gated recurrent units[J]. Renewable energy, 2020, 146: 760-768.
[17] 沈涛, 李舜酩. 针对滚动轴承故障的批标准化CNN-LSTM诊断方法[J]. 计算机集成制造系统, 2022, 28(12): 3946-3955.
SHEN T, LI S M.CNN-LSTM method with batch normalization for rolling bearing fault diagnosis[J]. Computer integrated manufacturing systems, 2022, 28(12): 3946-3955.
[18] 王进花, 韩金玉, 曹洁, 等. 基于AM和CNN的多级特征融合的风力发电机轴承故障诊断方法[J]. 太阳能学报, 2024, 45(5): 51-61.
WANG J H, HAN J Y, CAO J, et al.Fault diagnosis method of wind turbine bearing based on AM and CNN multi-level feature fusion[J]. Acta energiae solaris sinica, 2024, 45(5): 51-61.
[19] 谭启瑜, 马萍, 张宏立, 等. 基于图注意力网络的风力发电机齿轮箱故障诊断[J]. 太阳能学报, 2024, 45(1): 265-274.
TAN Q Y, MA P, ZHANG H L, et al.Fault diagnosis for wind turbine gearbox based on graph attention networks[J]. Acta energiae solaris sinica, 2024, 45(1): 265-274.
[20] 文靖杰, 王勇, 李金龙, 等. 多头自注意力机制的Faster R-CNN目标检测算法[J]. 现代电子技术, 2024, 47(7): 8-16.
WEN J J, WANG Y, LI J L, et al.Faster R-CNN object detection algorithm based on multi-head self-attention mechanism[J]. Modern electronics technique, 2024, 47(7): 8-16.
[21] 张玉彦, 张永奇, 孙春亚, 等. 不平衡样本下基于生成式对抗网络的风机叶片开裂状态识别[J]. 计算机集成制造系统, 2023, 29(2): 532-543.
ZHANG Y Y, ZHANG Y Q, SUN C Y, et al.Identification method of cracking state of wind turbine blade based on GAN under imbalanced samples[J]. Computer integrated manufacturing systems, 2023, 29(2): 532-543.
[22] PAPPAS C, KOVAIOS S, MORALIS-PEGIOS M, et al.Programmable Tanh-, ELU-, Sigmoid-, and Sin-based Nonlinear Activation Functions for Neuromorphic Photonics[J]. IEEE journal of selected topics in quantum electronics, 2023, 29(6): 6101210.
[23] CHUNG J Y, GULCEHRE C, CHO K-G, et al.Empirical evaluation of gated recurrent neural networks on sequence modeling[C]//In Proceedings of the 2014 Neural Information Processing Systems(NeurIPS 2014) Deep Learning and Representation Learning Workshop (NeurIPS 2014). Montréal, Canada, 2014.
[24] 熊红林, 樊重俊, 赵珊, 等. 基于多尺度卷积神经网络的玻璃表面缺陷检测方法[J]. 计算机集成制造系统, 2020, 26(4): 900-909.
XIONG H L, FAN C J, ZHAO S, et al.Glass surface defect detection method based on multiscale convolution neural network[J]. Computer integrated manufacturing systems, 2020, 26(4): 900-909.
[25] VASWANI A, SHAZEER N, PARMAR N, et al.Attention is all you need[C]//Advances in Neural Information Processing Systems 30 (NeurIPS 2017). Long Beach, California, USA, 2017: 5998-6008.