基于时空图卷积神经网络的强迫振荡定位与传播预测

Forced Oscillation Location and Propagation Prediction Based on Temporal Graph Convolutional Network

振荡源定位与传播预测是抑制强迫振荡和保证电力系统稳定的关键。现有方法未能充分利用电网的空间拓扑信息和振荡的时序特征,限制了定位和预测的精度。因此,该文提出一种基于时空图卷积神经网络的强迫振荡定位与传播预测方法。首先,根据节点特征和拓扑信息构建图数据,考虑到强迫振荡传播的快速性,通过切比雪夫多项式扩大节点空间感受野,提取振荡空间特征。同时,利用门控循环单元网络提取多个节点振荡数据的时序关联,通过时空图卷积单元融合空间和时序特征。然后,将定位与传播预测分别建模为分类和回归问题,训练时空图卷积神经网络模型。算例分析表明,所提方法具有更高的准确率,且在噪声和部分节点数据缺失的情况下依然具有较好的性能。

Oscillation source location and propagation prediction are the keys to suppressing the forced oscillation and ensuring the power system stability.Existing methods fail to fully utilize the spatial topology information of the power grid and the temporal characteristics of the oscillations,which limits the accuracy of location and prediction.Therefore,a forced oscillation location and propagation prediction method based on temporal graph convolutional network is proposed.Firstly,the graph data is constructed according to node features and topology information.Considering the rapidity of forced oscillation propagation,the spatial receptive field of nodes is expanded by Chebyshev polynomials,and the spatial features of forced oscillations are extracted.Meanwhile,the gated recurrent unit network is used to extract the temporal correlation of the oscillation data of multiple nodes.The spatial and temporal features are fused through the spatiotemporal graph convolution unit.Then,the location and propagation prediction are modeled as classification and regression problems,respectively,and temporal graph convolutional network models are trained.The case analysis shows that the method proposed in this paper has higher accuracy and still has good performance in the case of data noise and missing data of some nodes.