Short-Term Wind Power Prediction Method Based on Multimodal Feature Extraction-Convolutional Neural Network-Long-Short Term Memory Network

doi:10.12096/j.2096-4528.pgt.23101

Abstract

Abstract:

Objectives Weather and random factors can alter the statistical characteristics of errors. Therefore, this study considers feature extraction of various climate factors that affect wind power. To optimize the extraction of power time series features, a wind power prediction method based on multi-feature extraction (MFE), convolutional neural network (CNN), and long short-term memory (LSTM) network is proposed. Methods Firstly, 11 statistical features are extracted from numerical weather prediction (NWP) data. By extracting basic and statistical features, the original data is clustered, and prediction models are established according to categories to improve the adaptability of prediction models. Next, the network architecture of LSTM is improved. By leveraging the feature extraction ability of CNN and the nonlinear sequence prediction ability of LSTM, the historical information of wind power and NWP data is thoroughly explored. Finally, using the data from a wind farm in Xinjiang, China, the effectiveness and advantages of the proposed short-term wind power prediction method are verified by MFE and CNN ablation experiments. Results The MFE-CNN-LSTM prediction method shows a decrease in both root mean square error and mean absolute error, compared with the autoregressive integrated moving average (ARIMA), fully recurrent neural network (FRNN), MFE-LSTM and CNN-LSTM models. Conclusions The MFE-CNN-LSTM prediction method can effectively extract features, and MFE and CNN effectively improve prediction accuracy.

Key words: multi-feature extraction, convolutional neural network, long short-term memory networks, k-means clustering algorithm, wind power prediction, short-term prediction, ablation experiment

CLC Number:

TK 81

Honghai KUANG, Qian GUO. Short-Term Wind Power Prediction Method Based on Multimodal Feature Extraction-Convolutional Neural Network-Long-Short Term Memory Network[J]. Power Generation Technology, 2025, 46(1): 93-102.

Figures/Tables 15

Tab. 1 Normalized variables

类型	参数	意义
输入变量	V_{i b}	第i天高度b m处的风速(b=10, 30, 50, 70)
	D_{i b,1}	第i天高度b m处的风向角度的正弦值(b=10, 30, 50, 70)
	D_{i b,2}	第i天高度b m处的风向角度的余弦值(b=10, 30, 50, 70)
	Tⁱ	第i天的温度
	Sⁱ	第i天的湿度
	Hⁱ	第i天的压力
输出变量	Pⁱ	第i天的风电功率

Tab. 2 Basic characteristic variables

变量	意义
$v b, 1 i$	第i天高度b m处的最小风速(b=10, 30, 50, 70)
$v b, 2 i$	第i天高度b m处的最大风速(b=10, 30, 50, 70)
$d ¯ b, 1 i$	第i天高度b m处的风向角度的正弦值的平均值(b =10, 30,50, 70)
$d ¯ b, 2 i$	第i天高度b m处的风向角度的余弦值的平均值(b =10, 30,50, 70)
$t ¯$ ⁱ	第i天温度的平均值
$s ¯$ ⁱ	第i天湿度的平均值
$h ¯$ ⁱ	第i天压力的平均值

Tab. 2 Basic characteristic variables

变量	意义
$v b, 1 i$	第i天高度b m处的最小风速(b=10, 30, 50, 70)
$v b, 2 i$	第i天高度b m处的最大风速(b=10, 30, 50, 70)
$d ¯ b, 1 i$	第i天高度b m处的风向角度的正弦值的平均值(b =10, 30,50, 70)
$d ¯ b, 2 i$	第i天高度b m处的风向角度的余弦值的平均值(b =10, 30,50, 70)
$t ¯$ ⁱ	第i天温度的平均值
$s ¯$ ⁱ	第i天湿度的平均值
$h ¯$ ⁱ	第i天压力的平均值

Fig. 1 Feature clustering results

Fig. 2 Average value curves of three power types

Fig. 3 CNN network design

Fig. 4 LSTM network design

Fig. 5 Diagram of MFE-CNN-LSTM model prediction framework

Fig. 6 Diagram of MFE-CNN-LSTM structure

Fig. 7 Variation curve of loss function

Tab. 3 Multi-feature extraction ablation errors

类别	模型	e_MSE/MW	e_RMSE/MW	e_MAE/MW	e_MRE/ %
1	CNN-LSTM	3.626 8	1.904 4	1.460 4	15.312
1	MFE-CNN-LSTM	3.575 5	1.890 9	1.338 2	11.663
2	CNN-LSTM	3.710 1	1.926 2	1.499 6	17.413
2	MFE-CNN-LSTM	3.651 9	1.911 0	1.479 8	16.747
3	CNN-LSTM	6.500 5	2.549 6	1.732 2	20.924
3	MFE-CNN-LSTM	6.389 6	2.527 8	1.641 9	17.332

Fig. 8 Multi-feature extraction ablation experiments

Tab. 4 CNN network ablation errors

类别	模型	e_MSE/MW	e_RMSE/MW	e_MAE/MW	e_MRE/%
1	MFE-LSTM	3.692 3	1.921 5	1.415 1	13.806
1	MFE-CNN-LSTM	3.575 5	1.890 9	1.338 2	11.663
2	MFE-LSTM	3.869 1	1.967 0	1.466 4	13.951
2	MFE-CNN-LSTM	3.651 9	1.911 0	1.479 8	16.747
3	MFE-LSTM	6.865 4	2.620 2	1.713 4	21.924
3	MFE-CNN-LSTM	6.389 6	2.527 8	1.641 9	17.332

Fig. 9 CNN network ablation experiments

Fig. 10 Comparison of prediction results for different prediction models

Tab. 5 Comparison of errors of different wind power prediction models

类别	模型	e_MSE/ MW	e_RMSE/ MW	e_MAE/ MW	e_MRE/ %
1	ARIMA	5.043 8	2.245 8	1.711 2	17.350
	FRNN	3.925 9	1.981 4	1.550 0	16.979
	LSTM	3.754 9	1.937 8	1.465 3	15.784
	CVMD-SE-MCC-LSTM	3.672 5	1.916 4	1.500 0	13.154
	MFE-CNN-LSTM	3.575 5	1.890 9	1.338 2	11.663
2	ARIMA	4.405 7	2.099 0	1.613 6	13.061
	FRNN	3.856 1	1.963 7	1.575 9	17.328
	LSTM	3.952 9	1.988 2	1.579 7	16.878
	CVMD-SE-MCC-LSTM	3.787 2	1.946 1	1.494 7	12.128
	MFE-CNN-LSTM	3.651 9	1.911 0	1.479 8	16.747
3	ARIMA	8.418 5	2.901 5	1.956 1	22.777
	FRNN	7.360 8	2.713 1	1.883 4	25.111
	LSTM	7.012 9	2.648 2	1.812 9	21.634
	CVMD-SE-MCC-LSTM	6.662 9	2.581 3	1.955 3	23.920
	MFE-CNN-LSTM	6.389 6	2.527 8	1.641 9	17.332

References 27

1	张丽英，叶廷路，辛耀中，等．大规模风电接入电网的相关问题及措施[J]．中国电机工程学报，2010，30(25)：1-9．
	ZHANG L Y， YE T L， XIN Y Z，et al ．Problems and measures of power grid accommodating large scale wind power[J]．Proceedings of the CSEE，2010，30(25)：1-9．
2	雷旭，马鹏飞，宋智帅，等．计及风电预测误差的柔性负荷日内调度模型[J]．发电技术，2022，43(3)：485-491． doi:10.12096/j.2096-4528.pgt.20083
	LEI X， MA P F， SONG Z S，et al ．A flexible intraday load dispatch model considering wind power prediction errors[J]．Power Generation Technology，2022，43(3)：485-491． doi:10.12096/j.2096-4528.pgt.20083
3	叶瑞丽，郭志忠，刘瑞叶，等．基于风电功率预测误差分析的风电场储能容量优化方法[J]．电力系统自动化，2014，38(16)：28-34．
	YE R L， GUO Z Z， LIU R Y，et al ．A method for designing optimal energy storage system based on analysis of wind power forecast error[J]．Automation of Electric Power Systems，2014，38(16)：28-34．
4	王伟胜，王铮，董存，等．中国短期风电功率预测技术现状与误差分析[J]．电力系统自动化，2021，45(1)：17-27． doi:10.7500/AEPS20200324003
	WANG W S， WANG Z， DONG C，et al ．Status and error analysis of short-term forecasting technology of wind power in China[J]．Automation of Electric Power Systems，2021，45(1)：17-27． doi:10.7500/AEPS20200324003
5	王绍敏，王守相，赵倩宇，等．基于频域分解和精度加权集成的分布式风电功率预测方法[J]．电力建设，2023，44(5)：84-93．
	WANG S M， WANG S X， ZHAO Q Y，et al ．Distributed wind power forecasting method based on frequency domain decomposition and precision-weighted ensemble[J]．Electric Power Construction，2023，44(5)：84-93．
6	庞博文，丁月明，杜善慧，等．基于CEEMDAN-BO-LSTNet的风电出力短期预测[J]．电测与仪表，2023，60(9)：109-116．
	PANG B W， DING Y M， DU S H，et al ．Short-term forecasting of wind power output based on CEEMDAN-BO-LSTNet[J]．Electrical Measurement & Instrumentation，2023，60(9)：109-116．
7	蔡昌春，范靖浩，李源佳，等．基于TPA-MBLSTM模型的超短期风电功率预测[J]．电力科学与技术学报，2024，39(1)：47-56．
	CAI C C， FAN J H， LI Y J，et al ．Ultra-short-term wind power prediction based on TPA-MBLSTM model[J]．Journal of Electric Power Science and Technology，2024，39(1)：47-56．
8	苏向敬，朱敏轩，宇海波，等．基于频谱注意力和无交叉联合分位数回归的海上风电功率超短期概率预测[J]．电力系统保护与控制，2024，52(21)：103-116．
	SU X J， ZHU M X， YU H B，et al ．Ultra-short-term probability prediction of offshore wind power based on spectrum attention and cross-free joint quantile regression[J]．Power System Protection and Control，2024，52(21)：103-116．
9	胡宇晗，朱利鹏，李佳勇，等．融合深度误差反馈学习和注意力机制的短期风电功率预测[J]．电力系统保护与控制，2024，52(4)：100-108．
	HU Y H， ZHU L P， LI J Y，et al ．Short-term wind power forecasting based on deep error feedback learning and attention mechanism[J]．Power System Protection and Control，2024，52(4)：100-108．
10	LIU H， CHEN C ．Data processing strategies in wind energy forecasting models and applications：a comprehensive review[J]．Applied Energy，2019，249：392-408． doi:10.1016/j.apenergy.2019.04.188
11	杨国华，祁鑫，贾睿，等．基于CEEMD-SE的CNN&LSTM-GRU短期风电功率预测[J]．中国电力，2024，57(2)：55-61．
	YANG G H， QI X， JIA R，et al ．Short-term wind power forecast based on CNN&LSTM-GRU model integrated with CEEMD-SE algorithm[J]．Electric Power，2024，57(2)：55-61．
12	叶林，裴铭，路朋，等．基于天气分型的短期光伏功率组合预测方法[J]．电力系统自动化，2021，45(1)：44-54． doi:10.7500/AEPS20200613003
	YE L， PEI M， LU P，et al ．Combination forecasting method of short-term photovoltaic power based on weather classification[J]．Automation of Electric Power Systems，2021，45(1)：44-54． doi:10.7500/AEPS20200613003
13	张永蕊，阎洁，林爱美，等．多点数值天气预报风速和辐照度集中式修正方法研究[J]．发电技术，2022，43(2)：278-286． doi:10.12096/j.2096-4528.pgt.22005
	ZHANG Y R， YAN J， LIN A M，et al ．Integrated correction method of multi-point numerical weather prediction wind speed and irradiance[J]．Power Generation Technology，2022，43(2)：278-286． doi:10.12096/j.2096-4528.pgt.22005
14	任志玲，赵博雅．基于CNN预测的电厂热能联合循环控制研究[J]．控制工程，2019，26(8)：1544-1549．
	REN Z L， ZHAO B Y ．Research on power plant thermal energy combined cycle control based on CNN prediction[J]．Control Engineering of China，2019，26(8)：1544-1549．
15	赵申剑，黎彧君，符天凡，等．深度学习[M]．北京：人民邮电出版社，2017．
	ZHAO S J， LI Y J， FU T F，et al ．Deep learning[M]．Beijing：The People’s Posts and Telecommunications Press，2017．
16	董光德，李道明，陈咏涛，等．基于粒子群优化与卷积神经网络的电能质量扰动分类方法[J]．发电技术，2023，44(1)：136-142． doi:10.12096/j.2096-4528.pgt.22004
	DONG G D， LI D M， CHEN Y T，et al ．Power quality disturbance classification method based on particle swarm optimization and convolutional neural network[J]．Power Generation Technology，2023，44(1)：136-142． doi:10.12096/j.2096-4528.pgt.22004
17	于昕妍．微电网超短期负荷预测研究[D]．无锡：江南大学，2018．
	YU X Y ．Research on ultra-short-term load forecasting of microgrid[D]．Wuxi：Jiangnan University，2018．
18	WANG K， QI X， LIU H，et al ．Deep belief network based k-means cluster approach for short-term wind power forecasting[J]．Energy，2018，165：840-852． doi:10.1016/j.energy.2018.09.118
19	徐映梅，陈尧．季节ARIMA模型与LSTM神经网络预测的比较[J]．统计与决策，2021，37(2)：46-50．
	XU Y M， CHEN Y ．Comparison between seasonal ARIMA model and LSTM neural network forecast[J]．Statistics & Decision，2021，37(2)：46-50．
20	杨健，柳玉，黄坤鹏，等．考虑发电工况和站内损耗的风电场可用发电功率估算方法[J]．发电技术，2023，44(2)：235-243. doi:10.12096/j.2096-4528.pgt.21135
	YANG J， LIU Y， HUANG K P，et al ．A method for estimating available power of wind farms by considering the power generation conditions and station losses[J]．Power Generation Technology，2023，44(2)：235-243． doi:10.12096/j.2096-4528.pgt.21135
21	BOX G E P， COX D R ．An analysis of transformations[J]．Journal of the Royal Statistical Society：Series B，1964，26(2)：211-243． doi:10.1111/j.2517-6161.1964.tb00553.x
22	SALEH A E， MOUSTAFA M S， ABO-AL-EZ K M，et al ．A hybrid neuro-fuzzy power prediction system for wind energy generation[J]．International Journal of Electrical Power & Energy Systems，2016，74：384-395． doi:10.1016/j.ijepes.2015.07.039
23	ROSENSTEIN M T， COLLINS J J， DE LUCA C J ．A practical method for calculating largest Lyapunov exponents from small data sets[J]．Physica D：Nonlinear Phenomena，1993，65(1/2)：117-134． doi:10.1016/0167-2789(93)90009-p
24	WANG L， TIAN T， XU H，et al ．Short-term power load forecasting model based on t-SNE dimension reduction visualization analysis，VMD and LSSVM improved with chaotic sparrow search algorithm optimization[J]．Journal of Electrical Engineering & Technology，2022，17(5)：2675-2691． doi:10.1007/s42835-022-01101-7
25	彭晨宇，陈宁，高丙团．结合多重聚类和分层聚类的超短期风电功率预测方法[J]．电力系统自动化，2020，44(2)：173-180．
	PENG C Y， CHEN N， GAO B T ．Ultra-short-term wind power forecasting method combining multiple clustering and hierarchical clustering[J]．Automation of Electric Power Systems，2020，44(2)：173-180．
26	KLAMBAUER G， UNTERTHINER T， MAYR A，et al ．Self-normalizing neural networks[C]//Advances in Neural Information Processing Systems．Long Beach，USA：MIT Press，2017：971-980． doi:10.1016/j.toxlet.2017.07.175
27	赵卫东，董亮．机器学习[M]．北京：人民邮电出版社，2018． doi:10.1007/s10586-017-0843-2
	ZHAO W D， DONG L ．Machine learning[M]．Beijing：Posts & Telecom Press，2018. doi:10.1007/s10586-017-0843-2

[1]	Zhong LIU, Zehua ZHOU, Shuyun ZOU, Zhen LIU, Shuaicheng QIAO. Cavitation State Identification Method of Hydraulic Turbine Based on Knowledge Distillation and Convolutional Neural Network [J]. Power Generation Technology, 2025, 46(1): 161-170.
[2]	Zhan LIU, Yanyang BAO, Dazi LI. Fault Diagnosis Method of Wind Turbines Based on Wide Deep Convolutional Neural Network With Resampling and Principal Component Analysis [J]. Power Generation Technology, 2023, 44(6): 824-832.
[3]	Zhengling YANG, Ruxue WAGN, Jian QIAO, Xi ZHANG, Zhao YANG, Jun ZHANG. Analysis of the Influence of Atmospheric Pressure Difference on Spatial Correlation Prediction of Wind Speed [J]. Power Generation Technology, 2020, 41(6): 617-624.
[4]	Xincheng JIN,Xiuyuan YANG. Data Pre-processing Method Based on Distorted Data Noise Reduction and Its Application in Wind Power Prediction [J]. Power Generation Technology, 2020, 41(4): 447-451.