基于多尺度卷积-残差网络的短期风电预测

doi:10.3969/j.issn.2097-0706.2025.06.001

摘要/Abstract

摘要：

短期风电预测可为电力系统调度提供依据，防止风电波动对电力系统产生冲击。为解决现有用于短期风电预测的深度学习模型预测精度低的问题，考虑优化模型结构，提出基于多尺度卷积-残差网络的短期风电预测方法。所提出的多尺度卷积-残差网络具有特征提取尺度全面和稳定性强的特点。多尺度卷积部分并行使用了卷积核大小分别为3×3，5×5，7×7和9×9的卷积层，同时提取输入数据的局部细节和全局信息，残差部分引入跳跃连接，解决了卷积神经网络中梯度消失的问题。使用纳塔尔378 d数据集仿真结果表明，所提出方法能够实现对未来24 h的风力发电功率的准确预测。比DarkNet19，InceptionResNetV2，InceptionV3，ResNet18，ResNet50，ShuffleNet和Xception这7种对比算法的均方误差小43.55%以上。

关键词: 多尺度卷积, 残差, 深度学习, 优化模型结构, 短期风电预测

Abstract:

Short-term wind power forecast can provide a basis for power system scheduling and prevent power systems from severe impacts of wind power fluctuations. To improve the low prediction accuracy of existing deep learning models applied for short-term wind power prediction，a short-term wind power prediction method based on multiscale convolution-residual network is proposed to optimize the models. The proposed multiscale convolution-residual network is characterized by full range of feature extraction scales and strong stability，and the multiscale convolution part taking layers with convolutional kernel sizes of 3×3，5×5，7×7 and 9×9 is used to extracted detailed information and global information from the input data. By introducing hopping connections to the residual block，the vanishing gradient problem in the convolutional neural network is solved. The results of the simulation applying on the Natal 378-day dataset show that，the multiscale convolution-residual network can make an accurate prediction on wind power for the next 24 h，and the mean square error of the proposed network is more than 43.55% smaller than that of DarkNet19，InceptionResNetV2，InceptionV3，ResNet18，ResNet50，ShuffleNet and Xception.

Key words: multiscale convolution, residual, deep learning, model optimization, short-term wind power forecast

中图分类号:

TK89：TM614

殷林飞, 仝博文, 李雯吉. 基于多尺度卷积-残差网络的短期风电预测[J]. 综合智慧能源, 2025, 47(6): 1-10.

YIN Linfei, TONG Bowen, LI Wenji. Multiscale convolution-residual network for short-term wind power forecasting[J]. Integrated Intelligent Energy, 2025, 47(6): 1-10.

图/表 17

图1

图2

图3

图4

图5

图6

表1

图7

图8

表2

图9

图10

图11

表3

表4

表5

表6

参考文献 26

[1]	江善和, 李伟, 徐小艳, 等. 基于变分模态分解改进生成对抗网络的短期风电功率预测[J]. 综合智慧能源, 2024, 46(2): 28-35.
	JIANG Shanhe, LI Wei, XU Xiaoyan, et al. Short-term wind power forecasting based on variational mode decomposition and generative adversarial networks[J]. Integrated Intelligent Energy, 2024, 46(2): 28-35.
[2]	郑真, 朱峰, 马小丽, 等. 基于TL-LSTM的新能源功率短期预测[J]. 综合智慧能源, 2023, 45(1): 41-48.
	ZHENG Zhen, ZHU Feng, MA Xiaoli, et al. Short-term new energy power prediction based on TL-LSTM[J]. Integrated Intelligent Energy, 2023, 45(1): 41-48.
[3]	吴肃霜, 蔡小路, 李俊贤. 风电功率预测方法与最新技术发展研究[J]. 水电与新能源, 2024, 38(3):38-41,45.
	WU Sushuang, CAI Xiaolu, LI Junxian. On the wind power prediction methods and the latest technological development[J]. Hydropower and New Energy, 2024, 38(3):38-41,45.
[4]	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 Yuhan, ZHU Lipeng, LI Jiayong, et al. Short-term wind power forecasting with the integration of a deep error feedback learning and attention mechanism[J]. Power System Protection and Control, 2024, 52(4): 100-108.
[5]	殷林飞, 蒙雨洁. 基于DenseNet卷积神经网络的短期风电预测方法[J]. 综合智慧能源, 2024, 46(7):12-20.
	YIN Linfei, MENG Yujie. Short-term wind power forecasting based on DenseNet convolutional neural networks[J]. Integrated Intelligent Energy, 2024, 46(7):12-20.
[6]	CHEN H. A novel wind model downscaling with statistical regression and forecast for the cleaner energy[J]. Journal of Cleaner Production, 2024, 434: 140217.
[7]	赵增保. 北方不同下垫面风电功率预测方法研究[D]. 兰州: 兰州大学, 2014.
	ZHAO Zengbao. Research on forecast of wind power on different underlying surface in north China[D]. Lanzhou: Lanzhou University, 2014.
[8]	彭曙蓉, 彭家宜, 杨云皓, 等. 基于时变深度前馈神经网络的风电功率概率密度预测[J]. 电力科学与技术学报, 2023, 38(3): 84-93.
	PENG Shurong, PENG Jiayi, YANG Yunhao, et al. Wind power probability density prediction based on time-variant deep feed-forward neural network[J]. Journal of Electric Power Science and Technology, 2023, 38(3): 84-93.
[9]	杨宏伟. 面向风电功率预测的微尺度天气预报卷积神经网络方法研究[D]. 北京: 华北电力大学, 2022.
	YANG Hongwei. Research on microscale weather prediction method of wind power prediction based on convolutional neural network[D]. Beijing: North China Electric Power University, 2022.
[10]	向勇, 王一军, 孔寅恪. 基于深度循环神经网络的大规模风力发电功率预测[J]. 资源信息与工程, 2021, 36(2): 136-139.
	XIANG Yong, WANG Yijun, KONG Yinke. Large-scale wind power prediction based on deep-cycle neural network[J]. Resource Information and Engineering, 2021, 36(2): 136-139.
[11]	黄泽, 毕贵红, 谢旭, 等. 基于MBI-PBI-ResNet的超短期光伏功率预测[J]. 电力系统保护与控制, 2024, 52 (2): 165-176.
	HUANG Ze, BI Guihong, XIE Xu, et al. Ultra-short-term PV power prediction based on MBI-PBI-ResNet[J]. Power System Protection and Control, 2024, 52(2): 165-176.
[12]	朱奇先, 梁浩鹏, 赵小强, 等. 基于残差神经网络的滚动轴承故障诊断方法[J]. 兰州理工大学学报, 2022, 48(3): 86-93.
	ZHU Qixian, LIANG Haopeng, ZHAO Xiaoqiang, et al. Rolling bearing fault diagnosis method based on residual neural network[J]. Journal of Lanzhou University of Technology, 2022, 48(3): 86-93.
[13]	张涛, 任伟, 魏刚, 等. 基于CNN和LSTM的风机轴承剩余寿命预测方法研究[J]. 电气技术与经济, 2023(9): 26-28.
	ZHANG Tao, REN Wei, WEI Gang, et al. Research on residual life prediction method of wind turbine bearing based on CNN and LSTM[J]. Electrical Equipment and Economy, 2023(9): 26-28.
[14]	于建均, 刘耕源, 于乃功, 等. 轻量化改进XYZNet的RGB-D特征提取网络[J]. 计算机应用研究, 2024, 41(2):616-622.
	YU Jianjun, LIU Gengyuan, YU Naigong. RGB-D feature extraction network based on lightweight improved XYZNet[J]. Application Research of Computers, 2024, 41(2): 616-622.
[15]	李博文, 刘进锋. 一种基于生成对抗网络的图像去雾算法[J]. 宁夏工程技术, 2024, 23(2):185-192.
	LI Bowen, LIU Jinfeng. An image dehazing algorithm based on generative adversarial networks[J]. Ningxia Engineering Technology, 2024, 23(2): 185-192.
[16]	江达伟, 董阳阳, 张立东, 等. 基于深度学习的空中目标威胁评估技术研究[J/OL]. 系统仿真学报,1-12(2024-04-10)[2024-07-02].https://doi.org/10.16182/j.issn1004731x.joss.23-1323.
	JIANG Dawei, DONG Yangyang, ZHANG Lidong, et al. Research on air target threat assessment technology based on deep learning[J/OL]. Journal of System Simulation,1-12(2024-04-10)[2024-07-02].https://doi.org/10.16182/j.issn1004731x.joss.23-1323.
[17]	刘赟. ReLU激活函数下卷积神经网络的不同类型噪声增益研究[D]. 南京: 南京邮电大学, 2023.
	LIU Yun. Study on different types of noise benefit of convolutional neural networks under ReLU activation function[D]. Nanjing: Nanjing University of Posts and Telecommunications, 2023.
[18]	康钦豪. 基于深度学习的瓦楞纸实时盘点检测算法研究[D]. 重庆: 重庆理工大学, 2024.
	KANG Qinhao. Research on real time inventory detection algorithm for corrugated paper based on deep learning[D]. Chongqing: Chongqing University of Technology, 2024.
[19]	施保华, 井任月, 杨超, 等. 用于变电站保护压板状态识别的增强YOLO网络[J]. 电力系统保护与控制, 2021, 49(23): 163-170.
	SHI Baohua, JING Renyue, YANG Chao, et al. Enhanced YOLO network for status recognition of a substation protection plate[J]. Power System Protection and Control, 2021, 49(23): 163-170.
[20]	刘建伟, 赵会丹, 罗雄麟, 等. 深度学习批归一化及其相关算法研究进展[J]. 自动化学报, 2020, 46(6): 1090-1120.
	LIU Jianwei, ZHAO Huidan, LUO Xionglin, et al. Research progress on batch normalization of deep learning and its related algorithms[J]. Acta Automatica Sinica, 2020, 46(6): 1090-1120.
[21]	贺婷, 周宁, 吴啸宇. 基于深度全连接神经网络的储层有效砂体厚度预测[J]. 吉林大学学报(地球科学版), 2023, 53(4): 1262-1274.
	HE Ting, ZHOU Ning, WU Xiaoyu. Thickness prediction of reservoir effective sand body by deep fully connected neural network[J]. Journal of Jilin University(Earth Science Edition), 2023, 53(4): 1262-1274.
[22]	SILVA D P E, SALLES J L F, FARDIN J F, et al. Measured and forecasted weather and power dataset for management of an island and grid-connected microgrid[J]. Data in Brief, 2021, 39: 107513.
[23]	李玉龙, 童舟波, 陈玉莹. 基于Stacking预测模型的变桨功率优化策略研究[J]. 水电与新能源, 2024, 38(7):15-20.
	LI Yulong, TONG Zhoubo, CHEN Yuying. On the pitch power optimization strategy for wind turbine units based on Stacking prediction model[J]. Hydropower and New Energy, 2024, 38(7):15-20.
[24]	刘凯龙. 面向风电机组技术改进效果评价的风电功率回归预测[D]. 北京: 北京交通大学, 2023.
	LIU Kailong. Wind power regression prediction for evaluating the effect of technological improvements on wind turbines[D]. Beijing: Beijing Jiaotong University, 2023.
[25]	赵小平, 韩涛, 李永增, 等. 配电网单相接地故障行波滤波方法研究[J]. 智慧电力, 2018, 46(2): 45-53.
	ZHAO Xiaoping, HAN Tao, LI Yongzeng, et al. Filtering of traveling wave with single phase to ground fault in distribution networks[J]. Smart Power, 2018, 46(2): 45-53.
[26]	SILVA D P E, SALLES J L F, FARDIN J F, et al. Management of an island and grid-connected microgrid using hybrid economic model predictive control with weather data[J]. Applied Energy, 2020, 278: 115581.

变量	说明
时间	当地时间
温度/℃	瞬时、最高、最低和预测温度
湿度/%	瞬时、最大和最小湿度
露点温度/℃	瞬时、最高和最低露点温度
压力/hPa	瞬时、最大和最小压力
风速/(m·s^-1)	瞬时和预测风速
风向/(°)	预测风向
水平总辐射/(kJ·m^-2)	瞬时和预测水平总辐射
短波辐射/(kJ·m^-2)	瞬时短波辐射
降水量/mm	瞬时降水量
功率/W	风力发电功率

算法	层数	连接数	可学习参数/MB
多尺度卷积-残差网络	57	66	51.9
DarkNet19	64	63	20.8
InceptionResNetV2	824	921	55.8
InceptionV3	315	349	23.8
ResNet18	71	78	11.7
ResNet50	177	192	25.6
ShuffleNet	172	187	1.4
Xception	170	181	22.9

数据预处理方法	MSE/W²	RMSE/W	MAE/W	MAPE	SMAPE
无	9 099.17	95.39	70.57	0.71	0.43
插值法	2 697.96	51.94	34.22	0.37	0.26

最大训练轮数	MSE/W²	RMSE/W	MAE/W	MAPE	SMAPE
10	2 827.88	53.18	34.84	0.38	0.26
20	3 205.86	56.62	39.43	0.43	0.29
30	2 697.96	51.94	34.22	0.37	0.26
40	2 804.08	52.95	36.57	0.39	0.28

批量	MSE/W²	RMSE/W	MAE/W	MAPE	SMAPE
4	3 762.62	61.34	42.51	0.44	0.30
5	2 697.96	51.94	34.22	0.37	0.26
6	8 746.13	93.52	64.73	0.66	0.39
7	10 369.80	101.83	73.29	0.75	0.45