针对传统机器学习方法难以应对风电功率的强波动特性、预测精度不足的问题,提出一种基于PM-iTransformer-TCN和贝叶斯优化的超短期风电功率预测方法。首先通过Perona-Malik算法对原始风电功率序列进行平滑预处理,抑制其中的随机性成分;其次通过iTransformer模型联合风速、风向、温度、气压等多组输入变量,获得风电功率的初步预测值;最后利用时间卷积网络（TCN）基于风电功率数据的波动特性,实现多步预测误差的实时追踪和补偿。将贝叶斯优化（BO）应用于训练过程,以实现最优参数组合的选取。采用中国甘肃某风电场的现场数据进行方法验证。将所提方法与多种基线机器学习模型进行对比,结果表明,所提方法在未来2 h和4 h预测的归一化均方根误差（NRMSE）分别为18.95%和21.64%,优于传统的机器学习方法,可显著提升超短期风电功率预测的准确性。
Abstract
A multi-step ultra-short-term wind power forecasting method based on PM-iTransformer-TCN and Bayesian optimization （BO） is proposed to address the limitations of traditional machine learning methods, which struggle with the strong fluctuations of wind power and demonstrate insufficient prediction accuracy. First, the original wind power data is preprocessed using the Perona-Malik algorithm to suppress random components. Next, the iTransformer model combines multiple input variables, including wind speed, wind direction, temperature, and air pressure, to generate preliminary predicted values of wind power. Finally, real-time tracking and correction of multi-step prediction errors are achieved using Temporal Convolutional Networks （TCNs）, tailored to the fluctuation characteristics of wind power data. BO is applied in the training process to select the optimal parameter combination. The proposed method is compared with several baseline machine learning models using on-site data from a wind farm in Gansu Province, China. The results indicate that the normalized root mean square errors of the forecasting results at future 2 h and 4 h are 18.95% and 21.64%, respectively, outperforming traditional machine learning methods and significantly enhancing the accuracy of ultra-short-term wind power forecasting.
关键词
风电功率 /
预测 /
风电场 /
Perona-Malik算法 /
iTransformer模型 /
时间卷积网络 /
贝叶斯优化
Key words
wind power /
forecasting /
wind farm /
Perona-Malik algorithm /
iTransformer model /
temporal convolutional network /
Bayesian optimization
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 朱国鹏, 向玲, 范文振, 等. 基于样本卷积交互网络的风电场集群短期功率预测[J]. 太阳能学报, 2025, 46(1): 158-167.
Zhu G P, Xiang L, Fan W Z, et al.Short-term power prediction for wind farm cluster based on sample convolution interaction network[J]. Acta Energiae Solaris Sinica, 2025, 46(1): 158-167.
[2] Li K, Shen R F, Wang Z G, et al.An efficient wind speed prediction method based on a deep neural network without future information leakage[J]. Energy, 2023, 267: 126589.
[3] 宋江涛, 崔双喜, 刘洪广. 基于二次分解NGO-VMD残差项与长短时记忆神经网络的超短期风功率预测[J]. 科学技术与工程, 2023, 23(6): 2428-2437.
Song J T, Cui S X, Liu H G.Ultra-short-term wind power prediction based on the quadratic decomposition NGO-VMD residual term and long short-term memory(LSTM)[J]. Science Technology and Engineering, 2023, 23(6): 2428-2437.
[4] Chen P Y, Pedersen T, Bak-Jensen B, et al.ARIMA-based time series model of stochastic wind power generation[J]. IEEE Transactions on Power Systems, 2010, 25(2): 667-676.
[5] Cong Z H, Yu Y C, Li L Y, et al.Wind power time series simulation model based on typical daily output processes and Markov algorithm[J]. Global Energy Interconnection, 2022, 5(1): 44-54.
[6] 彭小圣, 熊磊, 文劲宇, 等. 风电集群短期及超短期功率预测精度改进方法综述[J]. 中国电机工程学报, 2016, 36(23): 6315-6326, 6596.
Peng X S, Xiong L, Wen J Y, et al.A summary of the state of the art for short-term and ultra-short-term wind power prediction of regions[J]. Proceedings of the CSEE, 2016, 36(23): 6315-6326, 6596.
[7] 付振宇, 王文胤, 凌小明, 等. 基于长短期记忆及自适应Kriging子集模拟优化的风电功率预测方法[J]. 电网与清洁能源, 2024, 40(11): 147-154.
Fu Z Y, Wang W Y, Ling X M, et al.A wind power prediction method based on long short-term memory and adaptive Kriging-subset simulation optimization[J]. Power System and Clean Energy, 2024, 40(11): 147-154.
[8] 宋江涛, 崔双喜, 樊小朝, 等. 基于SGMD-SE与优化TCN-BiLSTM/BiGRU的超短期风功率预测[J]. 太阳能学报, 2024, 45(10): 588-596.
Song J T, Cui S X, Fan X C, et al.Ultra-short-term wind power prediction based on SGMD-SE and optimized TCN-BiLSTM/BiGRU[J]. Acta Energiae Solaris Sinica, 2024, 45(10): 588-596.
[9] 韩宇超, 同向前, 邓亚平. 基于概率密度估计与时序Transformer网络的风功率日前区间预测[J]. 中国电机工程学报, 2024, 44(23): 9285-9296.
Han Y C, Tong X Q, Deng Y P.Probabilistic distribution estimation and temporal transformer-based interval prediction in day-ahead wind power prediction[J]. Proceedings of the CSEE, 2024, 44(23): 9285-9296.
[10] Li N, Dong J, Liu L Y, et al.A novel EMD and causal convolutional network integrated with Transformer for ultra short-term wind power forecasting[J]. International journal of Electrical Power & Energy Systems, 2023, 154: 109470.
[11] Yu M, Niu D X, Gao T, et al.A novel framework for ultra-short-term interval wind power prediction based on RF-WOA-VMD and BiGRU optimized by the attention mechanism[J]. Energy, 2023, 269: 126738.
[12] Yang S X, Zhou J X, Gu X W, et al.A comprehensive framework of the decomposition-based hybrid method for ultra-short-term wind power forecasting with on-site application[J]. Energy, 2024, 313: 133911.
[13] Chen Y S, Yu S, Islam S, et al.Decomposition-based wind power forecasting models and their boundary issue: an in-depth review and comprehensive discussion on potential solutions[J]. Energy Reports, 2022, 8: 8805-8820.
[14] 许梦田, 王洪哲, 赵成萍, 等. 基于短期风功率预测的数据预处理算法研究[J]. 可再生能源, 2019, 37(1): 119-125.
Xu M T, Wang H Z, Zhao C P, et al.Research on data preprocessing policy based on short-term wind power prediction[J]. Renewable Energy Resources, 2019, 37(1): 119-125.
[15] Zhao Y W, Shi X H, Zhou B X, et al.Wind power fluctuation smoothing with BESS considering ultra-short-term prediction[C]//2019 IEEE Innovative Smart Grid Technologies-Asia (ISGT Asia). Chengdu, China, 2019: 4025-4030.
[16] Qu Z J, Hou X X, Li J, et al.Short-term wind farm cluster power prediction based on dual feature extraction and quadratic decomposition aggregation[J]. Energy, 2024, 290: 130155.
[17] Wang J J, Tang X D.An intensive decomposition integration paradigm for short-term wind power forecasting based on feature extraction and optimal weighted combination strategy[J]. Measurement, 2023, 223: 113811.
[18] Guo J B, He C J, Zhang X T.Nonlinear edge-preserving diffusion with adaptive source for document images binarization[J]. Applied Mathematics and Computation, 2019, 351: 8-22.
[19] Liu Y, Hu T G, Zhang H R, et al.iTransformer: inverted transformers are effective for time series forecasting[C]//International Conference on Learning Representations (ICLR). Vienna, Austria, 2024: 11116-11140.
[20] Lea C, Flynn M D, Vidal R, et al.Temporal convolutional networks for action segmentation and detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, HI, USA, 2017: 1003-1012.
[21] 冯易鑫, 彭辉, 罗威. 聚类分析-神经网络-贝叶斯优化联合识别复合材料参数研究[J]. 力学学报, 2024, 56(11): 3333-3350.
Feng Y X, Peng H, Luo W.Research on parameter identification of composite materials by combination of self-consistent cluster analysis, neural network and Bayesian optimization[J]. Chinese Journal of Theoretical and Applied Mechanics, 2024, 56(11): 3333-3350.
[22] Fan H J, Zhen Z, Liu N, et al.Fluctuation pattern recognition based ultra-short-term wind power probabilistic forecasting method[J]. Energy, 2023, 266: 126420.
[23] Banik A, Behera C, Sarathkumar T V, et al.Uncertain wind power forecasting using LSTM-based prediction interval[J]. IET Renewable Power Generation, 2020, 14(14): 2657-2667.
[24] Zha W T, Liu J, Li Y L, et al.Ultra-short-term power forecast method for the wind farm based on feature selection and temporal convolution network[J]. ISA Transactions, 2022, 129: 405-414.
[25] Dai X R, Liu G P, Hu W S.An online-learning-enabled self-attention-based model for ultra-short-term wind power forecasting[J]. Energy, 2023, 272: 127173.
基金
浙江省自然科学基金（LBMHQN26E060001）; 国家自然科学基金（52375135; 52305137）
PDF(2063 KB)
