针对传统的风功率预测方法难以捕捉风电功率序列在频域内的非平稳、非线性特征,且未能充分利用风力机间的空间相互关系等问题,提出一种基于信号分解技术进行频率增强,并融合时空信息的网络DSTNet实现准确的超短期风电功率预测。在时间信息处理方面,利用离散余弦变换将风电功率序列从时域转换到频域,再通过通道注意力机制进行频率增强,然后采用解码器提取时间特征;在空间信息处理方面,基于风电场各风力机的位置信息构建图神经网络模型,提取各风力机节点与邻近节点的空间特征;最后,通过融合时间特征与空间特征进行超短期风电功率预测。基于百度KDD CUP发布的空间动态风电功率预测数据集为算例分析,结果表明,所提方法在超短期风电功率预测的所有场景中,即10 min、1 h、4 h,均具有最高的预测精度和稳定性。相比于排名第二的方法,该文方法平均绝对误差分别减少29.75%、19.11%、8.09%,均方根误差分别降低28.22%、13.44%、6.96%,就预测稳定性而言,决定系数R2分别提升1.78%、1.68%以及2.41%。
Abstract
To tackle the limitations of conventional wind power prediction methods in capturing non-stationary and nonlinear features of wind power sequences in the frequency domain and underutilization of spatial interrelations among wind turbines, we propose a novel network named DSTNet. This network integrates advanced signal decomposition technology for frequency enhancement while simultaneously incorporating both temporal and spatial information, enabling highly accurate ultra-short-term wind power forecasting. In terms of temporal information processing, discrete cosine transform is used to convert the wind power sequence from the time domain to the frequency domain, followed by frequency enhancement through a channel attention mechanism. To fully capture temporal dependencies, a decoder is employed to extract the intricate temporal features from the enhanced frequency signals. On the spatial front, a graph neural network model is employed, constructed based on the geographical layout of wind turbines within the farm. This GNN captures spatial features by modeling the relationships between each turbine node and its neighbors. Finally, the temporal and spatial features are fused to generate ultra-short-term wind power predictions. By seamlessly combining these temporal and spatial features, DSTNet enables superior ultra-short-term wind power forecasting, spanning prediction horizons of 10 minutes, 1 hour, and 4 hours. Our method is evaluated on the spatial dynamic wind power prediction dataset released by Baidu KDD CUP, and the results show that DSTNet outperforms all other methods in terms of both prediction accuracy and stability. Specifically, compared to the second-best method, the mean absolute error （MAE） is reduced by 29.75%, 19.11%, and 8.09%, respectively, and the mean square error （MSE） is reduced by 28.22%, 13.44%, and 6.96%, respectively. Furthermore, in terms of prediction stability, the coefficient of determination （R2） is increased by 1.78%, 1.68%, and 2.41%, respectively.
关键词
风电并网 /
时空数据 /
图神经网络 /
时间序列分解 /
离散余弦变换
Key words
wind power integration /
spatio-temporal data /
graph neural networks /
time series decomposition /
discrete cosine transform
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 江国乾, 徐向东, 白佳荣, 等. 基于多粒度时间卷积网络的超短期风功率预测[J]. 太阳能学报, 2024, 45(5): 104-111.
JIANG G Q, XU X D, BAI J R, et al.Ultra-short-term wind power prediction based on multi-granularity temporal convolution network[J]. Acta energiae solaris sinica, 2024, 45(5): 104-111.
[2] WU H X, XU J H, WANG J M, et al.Autoformer: decomposition transformers with auto-correlation for long-term series forecasting[J]. Advances in neural information processing systems, 2021, 34: 22419-22430.
[3] 夏晓荣, 胡鹏飞, 王飞, 等. 基于小波变换与优化BP神经网络的超短期光伏发电功率预测[J]. 电网与清洁能源, 2024, 40(10): 159-166.
XIA X R, HU P F, WANG F, et al.Ultra-short-term photovoltaic power prediction based on wavelet transform and optimal BP neural networks[J]. Power system and clean energy, 2024, 40(10): 159-166.
[4] BENTSEN L Ø, WARAKAGODA N D, STENBRO R, et al.Spatio-temporal wind speed forecasting using graph networks and novel Transformer architectures[J]. Applied energy, 2023, 333: 120565.
[5] LIU H, TIAN H Q, LI Y F.An EMD-recursive ARIMA method to predict wind speed for railway strong wind warning system[J]. Journal of wind engineering and industrial aerodynamics, 2015, 141: 27-38.
[6] WANG J Z, HU J M.A robust combination approach for short-term wind speed forecasting and analysis: combination of the ARIMA, ELM, SVM and LSSVM forecasts using a GPR model[J]. Energy, 2015, 93: 41-56.
[7] NASCIMENTO E G S, DE MELO T A C, MOREIRA D M. A transformer-based deep neural network with wavelet transform for forecasting wind speed and wind energy[J]. Energy, 2023, 278: 127678.
[8] GHADIMI N, AKBARIMAJD A, SHAYEGHI H, et al.Two stage forecast engine with feature selection technique and improved meta-heuristic algorithm for electricity load forecasting[J]. Energy, 2018, 161: 130-142.
[9] CAI W L, LIANG Y X, LIU X G, et al.MSGNet: learning multi-scale inter-series correlations for multivariate time series forecasting[J]. Proceedings of the AAAI conference on artificial intelligence, 2024, 38(10): 11141-11149.
[10] 傅质馨, 王宝池, 刘皓明, 等. 基于风速-功率组合分解重构的海上风电功率预测[J]. 太阳能学报, 2024, 45(11): 418-426.
FU Z X, WANG B C, LIU H M, et al.Offshore wind power forecasting based on wind speed-power combination decompositon[J]. Acta energiae solaris sinica, 2024, 45(11): 418-426.
[11] YI K, ZHANG Q, FAN W, et al.Frequency-domain MLPs are more effective learners in time series forecasting[J]. Advances in neural information processing systems, 2023, 36: 76656-76679.
[12] JIANG M W, ZENG P Y, WANG K, et al.FECAM: frequency enhanced channel attention mechanism for time series forecasting[J]. Advanced engineering informatics, 2023, 58: 102158.
[13] ZHANG Q Y, CHEN J H, XIAO G, et al.TransformGraph: a novel short-term electricity net load forecasting model[J]. Energy reports, 2023, 9: 2705-2717.
[14] YU C Q, YAN G X, YU C M, et al.A multi-factor driven spatiotemporal wind power prediction model based on ensemble deep graph attention reinforcement learning networks[J]. Energy, 2023, 263: 126034.
[15] DEMOLLI H, DOKUZ A S, ECEMIS A, et al.Wind power forecasting based on daily wind speed data using machine learning algorithms[J]. Energy conversion and management, 2019, 198: 111823.
[16] 戴长春, 郑晓亮, 杨铖, 等. 基于改进DBO-BiLSTM-GRU的光伏发电功率预测方法研究[J]. 自动化应用, 2024(19): 89-93.
DAI C C, ZHENG X L, YANG C, et al.Research on photovoltaic power generation prediction method based on improved DBO-BiLSTM-GRU[J]. Automation application, 2024(19): 89-93.
[17] AL-ALIMI D, ALRASSAS A M, AL-QANESS M A A, et al. TLIA: time-series forecasting model using long short-term memory integrated with artificial neural networks for volatile energy markets[J]. Applied energy, 2023, 343: 121230.
[18] VASWANI A, SHAZEER N, PARMAR N, et al.Attention is all you need[C]//Advances in Neural Information Processing Systems, NeurIPS proceedings 2017.
Long Beach, USA, 2017.
[19] 李士哲, 王霄慧, 刘帅. 考虑多变量相关性改进的风电场Transformer中长期预测模型[J]. 智慧电力, 2024, 52(4): 62-68, 107.
LI S Z, WANG X H, LIU S.Improved Transformer medium and long term prediction model of wind farm considering multivariate correlation[J]. Smart power, 2024, 52(4): 62-68, 107.
[20] WANG Y, XU H H, SONG M M, et al.A convolutional Transformer-based truncated Gaussian density network with data denoising for wind speed forecasting[J]. Applied energy, 2023, 333: 120601.
[21] ZHOU H Y, ZHANG S H, PENG J Q, et al.Informer: beyond efficient transformer for long sequence time-series forecasting[J]. Proceedings of the AAAI conference on artificial intelligence, 2021, 35(12): 11106-11115.
[22] 叶林, 赵永宁. 基于空间相关性的风电功率预测研究综述[J]. 电力系统自动化, 2014, 38(14): 126-135.
YE L, ZHAO Y N.A review on wind power prediction based on spatial correlation approach[J]. Automation of electric power systems, 2014, 38(14): 126-135.
[23] 马志侠, 张林鍹, 巴音塔娜, 等. 基于自适应二次分解与CNN-BiLSTM的超短期风电功率预测[J]. 太阳能学报, 2024, 45(6): 429-435.
MA Z X, ZHANG L X, BA Y, et al.Ultra-short-term wind power prediction based on adaptive quadratic mode decomposition and CNN-BiLSTM[J]. Acta energiae solaris sinica, 2024, 45(6): 429-435.
[24] ZHAO Z N, YUN S N, JIA L Y, et al.Hybrid VMD-CNN-GRU-based model for short-term forecasting of wind power considering spatio-temporal features[J]. Engineering applications of artificial intelligence, 2023, 121: 105982.
[25] 黄坤鹏. 风电功率超短期预测及可用发电功率计算[D]. 北京: 北京交通大学, 2022.
HUANG K P.Ultra-short-term wind power forecasting and available wind power calculation[D]. Beijing: Beijing Jiaotong University, 2022.
[26] SCARSELLI F, GORI M, TSOI A C, et al.The graph neural network model[J]. IEEE transactions on neural networks, 2009, 20(1): 61-80.
[27] Baidu KDD Cup2022[EB/OL]. https://aistudio.baidu.com/competition/detail/152/0/introduction.
基金
国家自然科学基金面上项目（62373148）; 国家重点研发计划（2021YFB2601300）
PDF(1959 KB)
