基于双通道时序特征融合网络的叶片覆冰检测算法

doi:10.13229/j.cnki.jdxbgxb.20250245

摘要/Abstract

摘要：

风能作为一种清洁、可再生能源，在全球范围内得到了显著推广，然而风力涡轮机叶片结冰问题严重影响了设备的安全运行和效率提升。现有叶片覆冰检测算法易受数据不平衡或噪声影响，导致准确率降低，因此亟需一种准确、高效的叶片覆冰检测算法以降低风机运行风险。本文提出了一种基于双通道时序特征融合网络的叶片覆冰检测算法：首先，对采集与监视控制系统数据进行数据预处理，并将特征工程应用至SCADA数据中；其次，提出了一种融合长短时记忆网络和注意力机制的检测模型，利用LSTM充分学习风机状态的时间依赖关系，采用注意力机制动态调整不同时间步的权重。最后，提出基于双通道时序特征融合网络的叶片覆冰检测算法，捕捉特征随时间变化的动态趋势，引入多头注意力机制增强特征交互能力。实验结果表明：相较其他算法，本文算法检测准确率提升了10%~15%，在风机叶片覆冰检测上具有较好的性能。

关键词: 覆冰检测, 时序特征融合, 长短时记忆网络, 注意力机制

Abstract:

As a clean and renewable energy source， wind energy has been extensively promoted globally. However， the issue of wind turbine blade icing has severely impacted the safe operation and efficiency of the equipment. Existing blade icing detection algorithms are prone to being affected by data imbalance or noise， leading to reduced accuracy. Consequently， there is an urgent need for a precise and efficient ice detection algorithm to mitigate operational risks. In this paper， we propose an ice detection algorithm for wind turbine blades based on a dual-channel temporal feature fusion network. First， we preprocess Supervisory Control and Data Acquisition （SCADA） data and apply feature engineering to SCADA data. Secondly， a detection model combining Long Short-Term Memory Network （LSTM） and attention mechanism is proposed. LSTM is used to fully learn the time dependence of fan state， and the attention mechanism is used to dynamically adjust the weights of different time steps. Finally， a blade ice detection algorithm based on dual channel temporal feature fusion network is proposed. Dynamic trends of features over time can be captured. Multi-head attention mechanism is introduced to enhance feature interaction. The experimental results show that： compared with other algorithms， the detection accuracy of the proposed algorithm is improved by 10%-15%. It has good performance in the ice detection of fan blades.

Key words: Icing detection, Temporal feature fusion, Long short-term memory network, Attention mechanism

中图分类号:

TP274

周求湛,陈霄,汪锋,张雯,李琛,武慧南,邓琬超,刘萍萍. 基于双通道时序特征融合网络的叶片覆冰检测算法[J]. 吉林大学学报(工学版), 2026, 56(9): 2489-2501.

Qiu-zhan ZHOU,Xiao CHEN,Feng WANG,Wen ZHANG,Chen LI,Hui-nan WU,Wan-chao DENG,Ping-ping LIU. Blade ice detection algorithm based on dual channel temporal feature fusion network[J]. Journal of Jilin University(Engineering and Technology Edition), 2026, 56(9): 2489-2501.

图/表 18

图1

图2

图3

表1

图4

表2

图5

图6

图7

图8

图9

图10

图11

图12

表3

图13

表4

表5

参考文献 22

[1]	柯坚. 碳达峰碳中和战略的法治回应[J]. 河北大学学报: 哲学社会科学版, 2025, 50(1): 123-124.
	Ke Jian. The rule of law response to the Carbon Peak Carbon neutral strategy[J]. Journal of Hebei University (Philosophy and Social Science),2025, 50(1):123-124.
[2]	安祥飞, 楚延锋. 风力发电机叶片防覆冰涂层研究进展[J]. 河南科技, 2024, 51(15): 76-80.
	An Xiang-fei, Chu Yan-feng. Research progress in anti-icing coating for wind turbine blades[J]. Henan Science and Technology, 2024, 51(15): 76-80.
[3]	余思锐, 宋孟杰, 沈俊, 等. 低温静止与运动表面结冰特性预测技术研究进展[J]. 哈尔滨工业大学学报, 2024, 56(10): 150-168.
	Yu Si-rui, Song Meng-jie, Shen Jun, et al. Research progress on icing characteristic prediction technologies for low temperature stationary and moving surfaces[J]. Journal of Harbin Institute of Technology, 2024, 56(10): 150-168.
[4]	Pérez J M P, Márquez F P G, Hernández D R. Economic viability analysis for icing blades detection in wind turbines[J]. Journal of Cleaner Production, 2016, 135: 1150-1160.
[5]	Jiang G Q, Yue R X, He Q, et al. Imbalanced learning for wind turbine blade icing detection via spatio-temporal attention model with a self-adaptive weight loss function[J]. Expert Systems with Applications, 2023, 229: 120428.
[6]	Guo J C, Song X W, Tang S F, et al. Fault diagnosis of wind turbine blade icing based on feature engineering and the PSO-ConvLSTM-transformer[J]. Ocean Engineering, 2024, 302: 117726.
[7]	Tian W W, Cheng X, Li G Y, et al. A multilevel convolutional recurrent neural network for blade icing detection of wind turbine[J]. IEEE Sensors Journal, 2021, 21(18): 20311-20323.
[8]	Chen W H, Cheng L S, Chang Z P, et al. Wind turbine blade icing detection using a novel bidirectional gated recurrent unit with temporal pattern attention and improved coot optimization algorithm[J]. Measurement Science and Technology, 2023, 34(1): 014004.
[9]	Jiang G Q, Li W Y, Bai J R, et al. SCADA data-driven blade icing detection for wind turbines: an enhanced spatio-temporal feature learning approach[J]. Measurement Science and Technology, 2023, 34(5): 054004.
[10]	Wang L, He Y G, Zhou Y Z, et al. A novel approach to wind turbine blade icing detection with limited sensor data via spatiotemporal attention Siamese network[J]. IEEE Transactions on Industrial Informatics, 2024,20(6): 8993-9005.
[11]	张好雨. 基于SCADA数据的风机叶片覆冰故障检测与预测[D]. 北京: 北京交通大学机械与电子控制工程学院, 2022.
	Zhang Hao-yu. Detection and prediction of wind turbine blade icing fault based on SCADA data[D]. Beijing: College of Mechanical and Electronic Control Engineering, Beijing Jiaotong University, 2022.
[12]	单锐, 杨婧, 朱文元, 等. 不同缺失比例下的缺失值插补方法比较[J]. 信息技术, 2023 (12): 52-56.
	Shan Rui, Yang Jing, Zhu Wen-yuan, et al. Comparison of missing value interpolation methods under different missing ratios[J]. Information Technology, 2023 (12): 52-56.
[13]	Chawla N V, Bowyer K W, Hall L O, et al. SMOTE: synthetic minority over-sampling technique[J]. Journal of Artificial Intelligence Research, 2002, 16: 321-357.
[14]	Puth M T, Neuhäuser M, Ruxton G D. Effective use of Pearson's product-moment correlation coefficient[J]. Animal Behaviour, 2014, 93: 183-189.
[15]	Ke G L, Meng Q, Finley T, et al. Lightgbm: a highly efficient gradient boosting decision tree[C]∥ NIPS'17: Proceedings of the 31st International Conference on Neural Information Processing Systems, New Orleans, USA, 2017.
[16]	Parent O, Ilinca A. Anti-icing and de-icing techniques for wind turbines: Critical review[J]. Cold Regions Science and Technology, 2011, 65(1): 88-96.
[17]	Skrimpas G A, Kleani K, Mijatovic N, et al. Detection of icing on wind turbine blades by means of vibration and power curve analysis[J]. Wind Energy, 2016, 19(10): 1819-1832.
[18]	胡梦婷, 罗晨. 基于MCNN-LSTM和交叉熵损失函数的轴承故障诊断[J]. 制造技术与机床, 2024, (9): 16-22.
	Hu Meng-ting, Luo Chen. Bearing fault diagnosis based on MCNN-LSTM and cross entropy loss function[J]. Manufacturing Technology & Machine Tool,2024 (9): 16-22.
[19]	Zhang Y, Xing K S, Bai R X, et al. An enhanced convolutional neural network for bearing fault diagnosis based on time-frequency image[J]. Measurement, 2020, 157: 107667.
[20]	Chandra M A, Bedi S S. Survey on SVM and their application in image classification[J]. International Journal of Information Technology, 2021, 13(5): 1-11.
[21]	Choe D E, Kim H C, Kim M H. Sequence-based modeling of deep learning with LSTM and GRU networks for structural damage detection of floating offshore wind turbine blades[J]. Renewable Energy, 2021, 174: 218-235.
[22]	Lin L J, Liang Y C, Liu L, et al. Estimating PM2. 5 concentrations using the machine learning RF-XGBoost model in guanzhong urban agglomeration China[J]. Remote Sensing, 2022, 14(20): 14205239.

相关文章 15

[1]	江晟,路启,夏淼磊,王稼禹. 基于特征融合注意力的RGB-D语义分割算法[J]. 吉林大学学报(工学版), 2026, 56(9): 2435-2443.
[2]	刘昕晖,陈晫群,吕岩. 基于深度学习的工业故障诊断综述[J]. 吉林大学学报(工学版), 2026, 56(7): 1759-1779.
[3]	曹洁,陈志峰,王进花,陈莉. 基于扩散模型和DenseNet的少样本齿轮箱故障诊断方法[J]. 吉林大学学报(工学版), 2026, 56(7): 1787-1797.
[4]	石峰,钮鹏,樊敏. 基于Faster R-CNN算法的公路路基路面不均匀变形检测[J]. 吉林大学学报(工学版), 2026, 56(7): 1950-1957.
[5]	缪君,严杰,杜荣华,李磊,储珺. 基于双向特征融合的物体位姿估计方法[J]. 吉林大学学报(工学版), 2026, 56(2): 523-532.
[6]	周求湛,李新萌,沈皓庆子,武慧南,李媛媛,荣静,胡春华,刘萍萍. 基于注意力机制的不平衡数据的非侵入式负荷分解[J]. 吉林大学学报(工学版), 2026, 56(1): 239-246.
[7]	冯志刚,任梦媛,董冰,于明月. 基于多频带特征图和改进SqueezeNet的滚动轴承故障诊断[J]. 吉林大学学报(工学版), 2026, 56(1): 96-108.
[8]	霍震,金立生,华强,贺阳. 基于边缘特征引导的智能汽车语义分割方法[J]. 吉林大学学报(工学版), 2025, 55(9): 3032-3041.
[9]	庄珊娜,王君帅,白晶,杜京瑾,王正友. 基于三维卷积与自注意力机制的视频行人重识别[J]. 吉林大学学报(工学版), 2025, 55(7): 2409-2417.
[10]	冯志刚,王首起,于明月. 基于变分模态提取及轻量级网络的滚动轴承故障诊断[J]. 吉林大学学报(工学版), 2025, 55(6): 1883-1891.
[11]	薛雅丽,俞潼安,崔闪,周李尊. 基于级联嵌套U-Net的红外小目标检测[J]. 吉林大学学报(工学版), 2025, 55(5): 1714-1721.
[12]	张河山,范梦伟,谭鑫,郑展骥,寇立明,徐进. 基于改进YOLOX的无人机航拍图像密集小目标车辆检测[J]. 吉林大学学报(工学版), 2025, 55(4): 1307-1318.
[13]	才华,王玉瑶,付强,马智勇,王伟刚,张晨洁. 基于注意力机制和特征融合的语义分割网络[J]. 吉林大学学报(工学版), 2025, 55(4): 1384-1395.
[14]	李扬,李现国,苗长云,徐晟. 基于双分支通道先验和Retinex的低照度图像增强算法[J]. 吉林大学学报(工学版), 2025, 55(3): 1028-1036.
[15]	张兰芳,李根泽,刘婷宇,余博. 局部多车影响下跟驰行为机理及建模[J]. 吉林大学学报(工学版), 2025, 55(3): 963-973.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

参数	值
Batch size	12
学习率	0.001
Epoch	100
Dropout	0.2
优化器	Adam

环境特征	权重	风机特征	权重
环境温度	0.163	发电机转速	0.160
风俗	0.160	有功功率	0.156
风向角	0.156	变桨电机温度1	0.136
平均风向角	0.136	X方向加速度	0.070

模型名称	准确率	精确率	召回率	F1值
XGBoost	0.749	0.713	0.702	0.707
LSTM	0.751	0.738	0.748	0.743
SVM	0.758	0.764	0.784	0.774
LSTM+Attention	0.853	0.819	0.829	0.824
DTFF-Net	0.894	0.867	0.864	0.866

模型编号	模型名称	修改内容说明
D1	完整DTFF-Net	保留双通道、残差特征、多头注意力模块
D2	w/o Dual-Channel	将风机与环境特征合并输入单通道LSTM，移除通道区分建模结构
D3	w/o Attention	移除多头注意力模块，改为简单连接融合两通道输出特征
D4	w/o Residual Input	不使用风速-功率残差和风速-转速残差，直接使用原始功率和转速特征

模型编号	准确率	精确率	召回率	F1值
D1	89.4	86.7	86.4	86.6
D2	85.5	82.3	81.6	81.9
D3	84.8	81.9	80.7	81.3
D4	82.9	80.2	78.5	79.3