针对现有基于深度学习的风力机尾流建模在连续参数化表征方面受限、图神经网络在该任务中尚未得到充分探索的问题,提出一种基于Graph Transformer的尾流建模方法。首先,构建一个涵盖多种环境和控制参数的大规模尾流数据集,设计基于坐标点采样与距离优先的尾流图结构构建方法,生成包含尾流场坐标、环境和控制参数的节点特征矩阵,为神经网络的训练提供数据输入。接着,提出基于Graph Transformer的尾流建模算法GT-Wake,利用自注意力机制动态调整图结构中邻居节点的权重分配,提升对尾流特性的局部特征与全局特征的捕捉能力。实验表明,GT-Wake 在 RMSE、MSE、MAE 和 MRE等建模指标上分别为1.84×10-3、0.34×10-5、1.45×10-2和0.97×10-3,精度优于 MLP、CNN 和 GraphSAGE等方法,且在泛化能力和计算效率间实现了良好平衡。
Abstract
To address the challenges of achieving continuous parameterization and the limited application of graph neural networks in existing deep learning-based wind farm wake modeling, this study proposes a Graph Transformer-based wind farm wake model. Firstly, a large-scale wake dataset is constructed, encompassing diverse environmental and control parameters. A coordinate-based sampling and distance-priority strategy is employed to design the wake graph structure, generating node feature matrices that include wake field coordinates, environmental, and control parameters for neural network training. Subsequently, a Graph Transformer-based wake modeling algorithm, i.e., GT-Wake, is developed. In which the self-attention mechanism is leveraged to dynamically adjust the weight allocation of neighboring nodes in the graph structure, thereby enhancing the model’s ability to capture both local and global wake characteristics. Experimental results demonstrate that compared to MLP, CNN, and GraphSAGE, GT-Wake significantly improves modeling accuracy and achieves a favorable balance between generalization capability and computational efficiency.
关键词
风力机 /
尾流 /
风电场 /
深度学习
Key words
wind turbine /
wake /
wind farm /
deep learning
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 魏赏赏, 李智寒, 陈一凯, 等. 基于状态扩张输入输出动态模态分解的风力机尾流降阶模型[J]. 太阳能学报, 2024, 45(10): 580-587.
WEI S S, LI Z H, CHEN Y K, et al.Reduced-order wake model of wind turbines based on state expansion input-output dynamic mode decomposition[J]. Acta energiae solaris sinica, 2024, 45(10): 580-587.
[2] 张虎, 许昌, 魏赏赏, 等. 基于稀疏增强动力学模态分解的风力机尾流模型研究[J]. 太阳能学报, 2024, 45(7): 681-690.
ZHANG H, XU C, WEI S S, et al.Study on wake model of wind turbine based on sparsity promoting dynamic mode decomposition[J]. Acta energiae solaris sinica, 2024, 45(7): 681-690.
[3] 张萍, 李成诚, 韩烨, 等. 漂浮式海上风力机三维尾流模型研究[J]. 太阳能学报, 2024, 45(7): 612-617.
ZHANG P, LI C C, HAN Y, et al.Research on three-dimensional wake model of floating offshore wind turbine[J]. Acta energiae solaris sinica, 2024, 45(7): 612-617.
[4] KATIC I, HOJSTRUP J, JESEN N O.A simple model for cluster efficiency[C]//European Wind Energy Association Conference and Exhibition. Roman, Itdy, 1986.
[5] BASTANKHAH M, PORTÉ-AGEL F.A new analytical model for wind-turbine wakes[J]. Renewable energy, 2014, 70: 116-123.
[6] ALFONSI G.Reynolds-averaged navier-stokes equations for turbulence modeling[J]. Applied mechanics reviews, 2009, 62(4): 040802.
[7] SAGAUT P.Large Eddy Simulation for Incompressible Flows[M]. Berlin, Heidelberg: Springer, 2002.
[8] LECUN Y, BENGIO Y, HINTON G.Deep learning[J]. Nature, 2015, 521(7553): 436-444.
[9] LI R, ZHANG J C, ZHAO X W.Dynamic wind farm wake modeling based on a Bilateral Convolutional Neural Network and high-fidelity LES data[J]. Energy, 2022, 258: 124845.
[10] ZHANG Z X, SANTONI C, HERGES T, et al.Time-averaged wind turbine wake flow field prediction using autoencoder convolutional neural networks[J]. Energies, 2021, 15(1): 41.
[11] RAISSI M, PERDIKARIS P, KARNIADAKIS G E. Physics informed deep learning (part I): data-driven solutions of nonlinear partial differential equations[EB/OL].2017: arXiv: 1711.10561. https://arxiv.org/abs/1711.10561
[12] SUN S X, CUI S S, HE T, et al.An integrated deep neural network framework for predicting the wake flow in the wind field[J]. Energy, 2024, 291: 130400.
[13] WANG L Y, CHEN M, LUO Z H, et al.Dynamic wake field reconstruction of wind turbine through Physics-Informed Neural Network and Sparse LiDAR data[J]. Energy, 2024, 291: 130401.
[14] CORSO G, STARK H, JEGELKA S, et al.Graph neural networks[J]. Nature reviews methods primers, 2024, 4: 17.
[15] LI S Y, ZHANG M R, PIGGOTT M D.End-to-end wind turbine wake modelling with deep graph representation learning[J]. Applied energy, 2023, 339(C): 120928.
[16] HAMILTON W L, YING R, LESKOVEC J.Inductive representation learning on large graphs[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, California, USA, 2017: 1025-1035.
[17] BASTANKHAH M, PORTÉ-AGEL F.Experimental and theoretical study of wind turbine wakes in yawed conditions[J]. Journal of fluid mechanics, 2016, 806: 506-541.
[18] CRESPO A, HERNA´NDEZ J. Turbulence characteristics in wind-turbine wakes[J]. Journal of wind engineering and industrial aerodynamics, 1996, 61(1): 71-85.
[19] KING J, FLEMING P, KING R, et al.Control-oriented model for secondary effects of wake steering[J]. Wind energy science, 2021, 6(3): 701-714.
[20] SHI Y S, HUANG Z J, FENG S K, et al.Masked label prediction: unified message passing model for semi-supervised classification[C]//Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence. Montreal, Canada, 2021: 1548-1554.
基金
国家重点研发计划（2023YFB4203200）
PDF(7634 KB)
