基于多时距风速波动过程划分的高铁沿线风速预测

张颖超; 安然; 陈昕; 叶小岭; 熊雄

doi:10.19818/j.cnki.1671-1637.2025.03.024

基于多时距风速波动过程划分的高铁沿线风速预测

doi: 10.19818/j.cnki.1671-1637.2025.03.024

张颖超^{1, 2},
安然¹,
陈昕¹,
叶小岭^{1, 2, ,},
熊雄^{1, 2}

1.
南京信息工程大学自动化学院, 江苏南京 210044
2.
南京信息工程大学大气环境与装备技术协同创新中心, 江苏南京 210044

基金项目:

国家自然科学基金项目 42275156

国家自然科学基金项目 42205150

江苏省自然科学青年基金项目 BK20210661

详细信息

作者简介:
张颖超(1960-)，男，江苏沛县人，南京信息工程大学教授，从事高铁气象研究

通讯作者:
叶小岭(1964-)，女，河南新乡人，南京信息工程大学教授，工学硕士

中图分类号: U238
计量
- 文章访问数: 172
- HTML全文浏览量: 47
- PDF下载量: 20
- 被引次数: 0
出版历程
- 收稿日期: 2024-07-09
- 录用日期: 2025-04-02
- 修回日期: 2025-03-06
- 刊出日期: 2025-06-28

Wind speed prediction along high-speed railway based on multi-time-interval wind speed fluctuation process division

ZHANG Ying-chao^{1, 2},
AN Ran¹,
CHEN Xin¹,
YE Xiao-ling^{1, 2
, ,},
XIONG Xiong^{1, 2}

1.
School of Automation, Nanjing University of Information Science and Technology, Nanjing 210044, Jiangsu, China
2.
Collaborative Innovation Center of Atmospheric Environment and Equipment Technology, Nanjing University of Information Science and Technology, Nanjing 210044, Jiangsu, China

Funds:

National Natural Science Foundation of China 42275156

National Natural Science Foundation of China 42205150

Natural Science Foundation of Jiangsu Province BK20210661

More Information

Corresponding author: YE Xiao-ling (1964-), female, professor, master, xyz.nim@163.com

Article Text (Baidu Translation)

摘要

摘要: 为了提升高铁沿线风速预测精度，实现预测预警，为大风调度提供足够时间窗口，以保障高铁大风场景下的运营安全，提出了一种基于风速波动过程划分的高铁沿线多时距注意力深度回声风速预测方法(MT-RFVMD-OD-Fu-Attention-DeepESN)。利用高铁沿线风速监测站点秒级风速采样数据和3 min风速采样数据，使用改进后的变分模态分解(RF-VMD)将2种分辨率的风速信号进行分解，并重构为趋势分量和脉动分量。通过连续小波变化(CWT)进行频率分析，找出变化周期，将2个分量划分为长度相等的时间序列单元。利用单元内风速物理特征和K-shape融合的聚类算法对2组趋势分量进行过程划分，形成风速波动过程数据库。最后，设计相似度优化动态时间调整(Op-DTW)算法，利用该算法在波动过程数据库中匹配出相似度较高的2种分辨率风速时间序列片段作为训练集，输入到所设计的多时距注意力深度回声状态预测网络(Fu-Attention-DeepESN)。依托京沪高铁沿线上3个风速站点实测风速监测数据进行实验验证，并同现有流行的风速预测方法进行对比。分析结果表明：站点K1005、K1245和K1066风速预测的均方根误差(RMSE)分别为0.234、0.282、0.306，平均绝对百分比误差(MAPE)分别为2.76%、2.27%、2.99%，风速正向误差(PWSE)分别为0.008、0.021、0.034。所提出的方法与对比方法中最好的相比，评价指标RMSE、MAPE、PWSE平均降低了27.8%、34.6%、27.1%，表明该研究能够有效处理高铁沿线秒级风信号中的复杂和非线性模式，提高风速预测的准确性和适应性。
- 铁路风速预测 /
- 风速波动过程划分 /
- 多时距特征融合 /
- 信号重构 /
- 信号匹配
Abstract: To improve the accuracy of wind speed prediction along high-speed railway, realize prediction and warning, provide a sufficient time window for high-wind dispatch, and ensure the operation safety of the high-speed railway in high-wind scenarios, a wind speed prediction method for multi-temporal attention deep echo state network along the high-speed railway based on wind speed fluctuation process division (MT-RFVMD-OD-Fu-Attention-DeepESN) was proposed. By using sampling data of second-level and three-minute wind speed of wind speed monitoring stations along the high-speed railway, as well as the improved radio frequency variational mode decomposition (RF-VMD), the wind speed signals of two resolutions were decomposed and reconstructed into trend components and pulsation components. The frequency analysis was performed by continuous wavelet transform (CWT) to find the variation cycle and divide the two components into time series units of equal length. The clustering algorithm of physical characteristics of wind speed in the units and K-shape fusion was used to divide the two groups of trend components into processes to form a database for wind speed fluctuation processes. Finally, an algorithm for similarity-optimized dynamic time warping (Op-DTW) was designed. The algorithm was used to select wind speed time series fragments with high similarity at two resolutions in the fluctuation process database as training sets, which were input into the designed multi-temporal attention deep echo state prediction network (Fu-Attention-DeepESN). The experiment was verified by using wind speed monitoring data from three wind speed stations along the Beijing-Shanghai high-speed railway and compared with the existing popular methods for wind speed prediction. Analysis results show that the root mean square errors (RMSEs) of wind speed prediction in stations K1005, K1245, and K1066 were 0.234, 0.282, and 0.306, respectively, and the mean absolute percentage errors (MAPEs) were 2.76%, 2.27%, and 2.99%, respectively. The positive wind speed errors (PWSEs) were 0.008, 0.021, and 0.034, respectively. Compared with those of the best of the compared methods, the evaluation indices RMSE, MAPE, and PWSE of the proposed method are reduced by 27.8%, 34.6%, and 27.1% on average, respectively. This proves that this study can effectively handle the complex and non-linear patterns in the second-level wind signals along the high-speed railway and improve the prediction accuracy and adaptability.
- railway wind speed forecast /
- wind speed fluctuation process division /
- multi-time interval feature fusion /
- signal reconstruction /
- signal matching

HTML全文

图 1 多时距注意力深度回声状态预测网络

Figure 1. Multi-temporal attention deep echo state prediction network

下载: 全尺寸图片幻灯片

图 2 深度回声状态网络

Figure 2. Deep echo state network

下载: 全尺寸图片幻灯片

图 3 高铁沿线多时距注意力深度回声风速预测方法框架

Figure 3. Framework of wind speed prediction method for multi-temporal attention deep echo along the high-speed railway

下载: 全尺寸图片幻灯片

图 4 RF-VMD分解结果

Figure 4. RF-VMD decomposition results

下载: 全尺寸图片幻灯片

图 5 信号重构示意

Figure 5. Schematic of signal reconstruction

下载: 全尺寸图片幻灯片

图 6 基于连续小波分析的频率分析

Figure 6. Frequency analysis based on CWT

下载: 全尺寸图片幻灯片

图 7 聚类个数及对应的轮廓系数

Figure 7. Numbers of clusters and silhouette coefficients

下载: 全尺寸图片幻灯片

图 8 秒级风速波动单元聚类结果

Figure 8. Clustering results of second-level wind speed fluctuation units

下载: 全尺寸图片幻灯片

图 9 3 min时距风速波动单元聚类结果

Figure 9. Clustering results of 3-minute wind speed fluctuation units

下载: 全尺寸图片幻灯片

图 10 多时距相似度匹配

Figure 10. Multi-temporal similarity matching

下载: 全尺寸图片幻灯片

图 11 K1245、K1005和K1066站点预测结果对比

Figure 11. Comparison of prediction results of site K1245, K1005 and K1066

下载: 全尺寸图片幻灯片

表 1 模型参数设置

Table 1. Model parameter settings

实验方法	初始参数设置
ARIMA	自回归阶数：11；差分阶数：2；移动平均阶数：1
SVR	核函数：RBF；正则化参数：10；核系数：0.1
XGBOOST	学习率：0.01；最大深度：5；树数量：200
LSTM	隐藏层数：3；隐藏单元数：128；激活函数：tanh函数；迭代次数：100；Dropout：0.2
1D-CNN	卷积核数量：64；卷积核大小：5；池化层：2；全连接单元：32
EEMD-LSTM	EEMD-集合数量：100；噪声强度：0.05；分解层数：8；LSTM-隐藏层数：3；隐藏单元数：128；激活函数：tanh；迭代次数：100；Dropout：0.2
PCA-LSTM	PCA-主成分保留方差：95%;LSTM-隐藏层数：3；隐藏单元数：128；激活函数：tanh函数；迭代次数：100；Dropout：0.2
MW-LSTM	MW-小波函数：db4；分解层数：3层：LSTM-隐藏层数：3；隐藏单元数：128；激活函数：tanh函数；迭代次数：100；Dropout：0.2
VMD-GA-BP	VMD-模态数：8；惩罚因子：2 000；GA种群大小：50；迭代次数：100；BP网络层：64
VMD-PSO-BiLSTM	VMD模态数：7；PSO种群大小：30；迭代次数：50；BiLSTM单元数：64
DEEPER	隐藏层大小：64；层数：2；初始学习率：0.001；训练轮数：100
WINDFORMER	多头注意力的头数：4；隐藏单元数：256；初始学习率：0.001
CEEMDAN-WOA-SVR	CEEMDAN分解层数：8；WOA种群大小：20；迭代次数：50；核函数：RBF；正则化参数：10；核系数：0.1
Fu-Attention- DeepESN	DeepESN-储备池规模：300；层数：3；频谱半径0.9；泄露率：0.3；激活函数：tanh函数；多头注意力的头数：8；隐藏单元数：256；初始学习率：0.001
MT-RFVMD-OD-Fu- Attention-DeepESN本文方法	RF-VMD: 分解层数：7；DeepESN-储备池规模：300；层数：3；频谱半径0.9；泄露率：0.3；激活函数：tanh函数；多头注意力的头数：8；隐藏单元数：256；初始学习率：0.001

下载: 导出CSV

表 2 模型预测结果对比

Table 2. Comparison of model prediction results

实验方法	K1005站点			K1066站点			K1245站点
实验方法	MAPE/%	RMSE	PWSE	MAPE/%	RMSE	PWSE	MAPE/%	RMSE	PWSE
ARIMA	9.00	1.099	-0.682	7.32	1.150	-0.738	7.07	1.047	-0.636
SVR	8.56	0.936	2.008	7.07	1.102	-0.650	6.46	0.967	-0.431
XGBOOST	8.75	1.064	1.616	7.39	1.136	0.422	6.54	0.968	0.592
LSTM	8.54	0.872	-0.479	7.08	1.082	-0.534	7.92	0.827	-0.340
1D-CNN	8.95	0.893	-0.945	7.44	1.126	-0.862	6.56	0.906	-0.460
EEMD-LSTM	5.69	0.671	-0.258	3.20	0.628	-0.037	4.48	0.689	0.261
PCA-LSTM	7.07	0.777	-0.861	7.10	0.856	-0.795	4.71	0.714	-0.628
MW-LSTM	4.67	0.501	0.095	5.23	0.826	-0.441	3.33	0.522	-0.266
VMD-GA-BP	4.46	0.571	0.589	4.17	0.625	0.078	3.85	0.548	0.384
VMD-PSO-BiLSTM	4.44	0.555	-0.154	4.81	0.726	0.267	3.93	0.534	-0.219
DEEPER	6.71	0.719	0.578	5.33	0.840	0.282	4.11	0.596	0.354
WINDFORMER	5.71	0.630	0.232	4.43	0.680	0.044	3.16	0.486	0.351
CEEMDAN-WOA-SVR	3.38	0.796	-0.109	5.22	0.823	0.438	3.73	0.516	0.479
Fu-Attention-DeepESN	4.36	0.452	0.094	2.96	0.494	0.235	2.78	0.416	0.206
本文方法	2.76	0.234	0.008	2.99	0.306	0.034	2.27	0.282	0.021

下载: 导出CSV

表 3 消融实验评价指标

Table 3. Evaluation indices of ablation experiments

实验方法	K1005站点			K1066站点			K1245站点
实验方法	MAPE/%	RMSE	PWSE	MAPE/%	RMSE	PWSE	MAPE/%	RMSE	PWSE
MT-RFVMD-Fu-Attention-DeepESN	2.25	0.334	0.063	2.98	0.378	0.134	2.67	0.309	0.097
MT-VMD-OD-Fu-Attention-DeepESN	1.65	0.246	0.019	2.38	0.268	0.023	1.98	0.278	0.065
RFVMD-OD-Fu-Attention-DeepESN	2.87	0.423	-0.057	3.05	0.556	-0.103	3.01	0.368	-0.091
MT-RFVMD-OD-Fu-DeepESN	3.54	0.389	0.083	3.82	0.491	0.071	2.61	0.475	0.011
本文方法	2.76	0.234	0.008	2.99	0.306	0.034	2.27	0.282	0.021

下载: 导出CSV

参考文献(37)

[1]	李晓健, 陈雍君, 邱实, 等. 复杂地区铁路工程建设风险知识图谱的建立与分析方法[J]. 铁道学报, 2025, 47(5): 187-196. doi: 10.3969/j.issn.1001-8360.2025.05.020 LI Xiao-jian, CHEN Yong-jun, QIU Shi, et al. Establishment and analysis method for risk knowledge graph of railway construction in complex areas[J]. Journal of the China Railway Society, 2025, 47(5): 187-196. doi: 10.3969/j.issn.1001-8360.2025.05.020
[2]	DELAUNAY D, CLÉON L M, SACRÉ C, et al. Designing a wind alarm system for the TGV-Méditerranée[C]//SMITH D A, LETCHFORD C W. Proceedings of the 11th International Conference on Wind Engineering. Lubbock: Texas Tech University, 2005: 1-8.
[3]	HOPPMANN U, KOENIG S, TIELKES T, et al. A short-term strong wind prediction model for railway application: design and verification[J]. Journal of Wind Engineering and Industrial Aerodynamics, 2002, 90(10): 1127-1134.
[4]	包云, 王瑞, 王彤. 欧洲对高速铁路横风的研究[J]. 中国铁路, 2015(3): 8-11. doi: 10.3969/j.issn.1001-683X.2015.03.002 BAO Yun, WANG Rui, WANG Tong. Study on crosswind of high-speed railway in Europe[J]. Chinese Railways, 2015(3): 8-11. doi: 10.3969/j.issn.1001-683X.2015.03.002
[5]	温旭军. 高速铁路大风预警方法及预警系统的研究[D]. 北京: 北京交通大学, 2023. WEN Xu-jun. Research on early warning method and early warning system of high-speed railway gale[D]. Beijing: Beijing Jiaotong University, 2023.
[6]	BOUZGOU H, BENOUDJIT N. Multiple architecture system for wind speed prediction[J]. Applied Energy, 2011, 88(7): 2463-2471.
[7]	PEARRE N S, SWAN L G. Statistical approach for improved wind speed forecasting for wind power production[J]. Sustainable Energy Technologies and Assessments, 2018, 27: 180-191.
[8]	MEMARZADEH G, KEYNIA F. A new short-term wind speed forecasting method based on fine-tuned LSTM neural network and optimal input sets[J]. Energy Conversion and Management, 2020, 213: 112824.
[9]	JASEENA K U, KOVOOR B C. Decomposition-based hybrid wind speed forecasting model using deep bidirectional LSTM networks[J]. Energy Conversion and Management, 2021, 234: 113944.
[10]	FANTINI D G, SILVA R N, SIQUEIRA M B B, et al. Wind speed short-term prediction using recurrent neural network GRU model and stationary wavelet transform GRU hybrid model[J]. Energy Conversion and Management, 2024, 308: 118333.
[11]	JOSEPH L P, DEO R C, PRASAD R, et al. Near real-time wind speed forecast model with bidirectional LSTM networks[J]. Renewable Energy, 2023, 204: 39-58. doi: 10.3969/j.issn.1671-5292.2023.01.006
[12]	MA Z R, CHEN H W, WANG J J, et al. Application of hybrid model based on double decomposition, error correction and deep learning in short-term wind speed prediction[J]. Energy Conversion and Management, 2020, 205: 112345.
[13]	DUAN J K, ZUO H C, BAI Y L, et al. Short-term wind speed forecasting using recurrent neural networks with error correction[J]. Energy, 2021, 217: 119397.
[14]	ZHANG S H, WANG C, LIAO P, et al. Wind speed forecasting based on model selection, fuzzy cluster, and multi-objective algorithm and wind energy simulation by Betz's Theory[J]. Expert Systems with Applications, 2022, 193: 116509.
[15]	LV S X, WANG L. Multivariate wind speed forecasting based on multi-objective feature selection approach and hybrid deep learning model[J]. Energy, 2023, 263: 126100.
[16]	SHANG Z H, CHEN Y, CHEN Y H, et al. Decomposition-based wind speed forecasting model using causal convolutional network and attention mechanism[J]. Expert Systems with Applications, 2023, 223: 119878.
[17]	YAMAGUCHI A, ISHIHARA T. Maximum instantaneous wind speed forecasting and performance evaluation by using numerical weather prediction and on-site measurement[J]. Atmosphere, 2021, 12(3): 316.
[18]	金曈宇, 叶小岭, 熊雄, 等. 高铁沿线大风预测技术研究[J]. 铁道科学与工程学报, 2021, 18(3): 615-622. JIN Tong-yu, YE Xiao-ling, XIONG Xiong, et al. Research on prediction technology of high wind along high-speed railway[J]. Journal of Railway Science and Engineering, 2021, 18(3): 615-622.
[19]	何旭辉, 段泉成, 严磊. 基于DeepAR的短期风速概率预测[J]. 铁道学报, 2023, 45(7): 152-160. doi: 10.3969/j.issn.1001-8360.2023.07.018 HE Xu-hui, DUAN Quan-cheng, YAN Lei. Short-term wind speed probabilistic prediction model using DeepAR[J]. Journal of the China Railway Society, 2023, 45(7): 152-160. doi: 10.3969/j.issn.1001-8360.2023.07.018
[20]	LIU C, HE S B, LIU H Y, et al. WindTrans: Transformer-based wind speed forecasting method for high-speed railway[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(6): 4947-4963.
[21]	江星星, 宋秋昱, 朱忠奎, 等. 基于收敛趋势变分模式分解的齿轮箱故障诊断方法[J]. 交通运输工程学报, 2022, 22(1): 177-189. doi: 10.19818/j.cnki.1671-1637.2022.01.015 JIANG Xing-xing, SONG Qiu-yu, ZHU Zhong-kui, et al. Gearbox fault diagnosis method based on convergent trend-guided variational mode decomposition[J]. Journal of Traffic and Transportation Engineering, 2022, 22(1): 177-189. doi: 10.19818/j.cnki.1671-1637.2022.01.015
[22]	ZHANG Y G, PAN G F, CHEN B, et al. Short-term wind speed prediction model based on GA-ANN improved by VMD[J]. Renewable Energy, 2020, 156: 1373-1388.
[23]	LI Y T, WU Z Y, SU Y. Adaptive short-term wind power forecasting with concept drifts[J]. Renewable Energy, 2023, 217: 119146.
[24]	YANG L X, ZHANG Z J. A deep attention convolutional recurrent network assisted by K-shape clustering and enhanced memory for short term wind speed predictions[J]. IEEE Transactions on Sustainable Energy, 2022, 13(2): 856-867.
[25]	CHEN X, YE X L, XIONG X, et al. Improving the accuracy of wind speed spatial interpolation: A pre-processing algorithm for wind speed dynamic time warping interpolation[J]. Energy, 2024, 295: 130876.
[26]	EL AMOURI H, LAMPERT T, GANÇARSKI P, et al. Constrained DTW preserving shapelets for explainable time-series clustering[J]. Pattern Recognition, 2023, 143: 109804.
[27]	WANG G, SUN L F, WANG A J, et al. Lithium battery remaining useful life prediction using VMD fusion with attention mechanism and TCN[J]. Journal of Energy Storage, 2024, 93: 112330.
[28]	LI Z X, LI L W, CHEN J, et al. A multi-head attention mechanism aided hybrid network for identifying batteries' state of charge[J]. Energy, 2024, 286: 129504.
[29]	WANG B W, LUN S X, LI M, et al. Echo state network structure optimization algorithm based on correlation analysis[J]. Applied Soft Computing, 2024, 152: 111214.
[30]	GAO R B, LI R L, HU M H, et al. Dynamic ensemble deep echo state network for significant wave height forecasting[J]. Applied Energy, 2023, 329: 120261.
[31]	MA Q L, ZHENG Z J, ZHUANG W Q, et al. Echo Memory-augmented Network for time series classification[J]. Neural Networks, 2021, 133: 177-192.
[32]	曲栩, 甘锐, 安博成, 等. 基于广义时空图卷积网络的交通群体运动态势预测[J]. 交通运输工程学报, 2022, 22(3): 79-88. doi: 10.19818/j.cnki.1671-1637.2022.03.006 QU Xu, GAN Rui, AN Bo-cheng, et al. Prediction of traffic swarm movement situation based on generalized spatio-temporal graph convolution network[J]. Journal of Traffic and Transportation Engineering, 2022, 22(3): 79-88. doi: 10.19818/j.cnki.1671-1637.2022.03.006
[33]	VY V, LEE Y, BAK J, et al. Damage localization using acoustic emission sensors via convolutional neural network and continuous wavelet transform[J]. Mechanical Systems and Signal Processing, 2023, 204: 110831.
[34]	YAN Y, WANG X R, REN F, et al. Wind speed prediction using a hybrid model of EEMD and LSTM considering seasonal features[J]. Energy Reports, 2022, 8: 8965-8980.
[35]	GENG D W, ZHANG H F, WU H Y. Short-term wind speed prediction based on principal component analysis and LSTM[J]. Applied Sciences, 2020, 10(13): 4416.
[36]	YU C J, LI Y L, CHEN Q, et al. Matrix-based wavelet transformation embedded in recurrent neural networks for wind speed prediction[J]. Applied Energy, 2022, 324: 119692.
[37]	LI J L, SONG Z H, WANG X F, et al. A novel offshore wind farm typhoon wind speed prediction model based on PSO-Bi-LSTM improved by VMD[J]. Energy, 2022, 251: 123848.