风电场功率数据质量对风电预测具有重要意义.然而,由于人为操作、传感器故障、网络故障或通信拥堵等原因,风电场功率数据容易出现大面积缺失问题.因此,结合多头注意力机制(multi-head attention,MA)和图注意力网络(graph attention network,GAT)构建了MAGAT 模型,其中 GAT 层以异质图的方式刻画及提取风电场已知数据与缺失数据的关联关系,MA层挖掘风电场数据特征与缺失功率数据之间的映射关系,从而实现风电场功率缺失数据的高精度填充.在以我国江苏某风电场运行监测数据为对象的算例分析中,与其他先进填充算法相比,所提方法在不同缺失类型、不同缺失率等多个场景下均具有更好表现,表明所提方法在风电场缺失数据填充任务上的有效性及稳定性.
MAGAT-based Method for Imputing Power Loss Data in Wind Farms
Wind farms'quality of power data is significant for wind power forecasting.However,due to human error,sensor failures,network issues,or communication congestion,wind farm power data is prone to extensive missing values.Therefore,this article proposes a MAGAT(Multi-Head Attention and Graph Attention Network)model that combines Multi-head Attention(MA)and Graph Attention Network(GAT)to address this issue.Using a heterogeneous graph representation,the GAT layer is used to characterize and extract the associative relationships between known and missing data in the wind farm.On the other hand,the MA layer focuses on mining the mapping relationship between the features of wind farm data and the missing power data,enabling accurate filling of missing power data in the wind farm.The proposed method is compared to other advanced filling algorithms in an empirical analysis using operational monitoring data from a wind farm in Jiangsu,China.The results show that the proposed method outperforms the others in various scenarios,including different types and rates of missing data.This demonstrates the effectiveness and stability of the proposed method for filling in missing data in wind farms.
万方数据
维普
