考虑风电机组时空特性的风电场超短期功率预测模型

An Ultra-Short-Term Power Prediction Model for Wind Farms Considering the Spatiotemporal Characteristics of Wind Turbines

中文关键词:ultra-short-term power prediction  spatiotemporal characteristics  convolutional neural network  bidirectional gated recurrent unit

keywords:超短期功率预测  时空特性  卷积神经网络  双向门控循环单元

中文摘要:传统的风电功率预测统计模型在进行风电场超短期功率预测时，难以处理高维数据且忽略风电机组间时空相关性。针对这一问题，提出了考虑风电机组时空相关性的超短期预测模型。首先，通过快速动态时间扭曲算法度量机组的空间相似度，筛选出与目标风机相似的机组群作为模型输入数据。其次，采用基于深度学习的三维卷积神经网络和双向门控循环单元的混合模型，提取数据的时空相关性；同时引入自注意力机制和残差网络，增强模型对关键信息的提取。最后，以某风电场的实际运行数据为例，进行风机功率预测实验。实验结果表明，与现有的算法相比，所提方法预测误差更小，预测精度更高，拟合效果明显优于其他基础模型。

Abstract:Traditional statistical models for wind power prediction struggle to process high-dimensional data and neglect the spatiotemporal correlations between wind turbines when conducting ultra-short-term power prediction for wind farms. To solve this problem，an ultra-short-term power prediction model considering the spatiotemporal correlations of wind turbines is proposed in this paper. Firstly，the fast dynamic time warping （FastDTW） algorithm is adopted to measure the spatial similarity of wind turbines，and a cluster of wind turbines similar to the target one is selected as the input data for the model. Secondly，a hybrid deep learning model combining the three-dimensional convolutional neural network （3D CNN） and bidirectional gated recurrent unit （BiGRU） is adopted to extract the spatiotemporal correlations of the data； meanwhile，the self-attention mechanism and residual network （ResNet） are introduced to enhance the model's ability to extract key information. Finally，by taking the actual operational data of a wind farm as a case study，experiments on wind turbine power prediction are conducted. The experimental results show that compared with existing algorithms，the proposed method yields a smaller prediction error and a higher prediction accuracy，and it’s fitting performance is significantly superior to that of other basic models.