基于稳态特征量输入的大电网主导失稳机组辨识

Identification of Leading Instable Generators for Large-scale Power Grid Based on Steady-state Characteristic Inputs

以稳态特征量为输入的数据驱动稳定评估模型在新型电力系统安全防御系统中有重要的应用前景,但需要在模型设计中解决节点数量庞大和网络结构复杂带来的关键特征聚焦难题,并提供失稳模式等更为丰富的评估信息。因此,设计了一套基于稳态信息输入实现大电网主导失稳机群预测的深度学习稳定评估模型。首先,提出了一种异构的图和节点特征的动态池化降维模型,可伴随特征聚合过程,按节点特征相似性动态归并节点,实现大规模电网拓扑、节点数量和特征的并行降维。然后,提出了一种单机扫描型主导失稳机组分类器模型,通过全局注意力聚合将全网机组的相对运动信息集成到每台发电机特征向量中,使主导失稳机组辨识模型在结构上可以应对发电机组数量变化,具有很好的泛化能力。最后,在实际大规模电网中进行模型验证,并可视化地分析了关键环节的作用效果和应用性能。

The data-driven stability assessment model with the input of steady-state characteristics has an important application prospect in the safety and stability research and judgment of new power systems,but it needs to solve the problem of extracting key characteristics caused by the large number of nodes and complex network structure in the model design,and provide more abundant assessment information such as instability modes.Therefore,a set of deep learning stability assessment model based on steady-state information input is designed for prediction of the leading instable generators of large-scale power grid.Firstly,a dynamic pooling dimensionality reduction model of heterogeneous graphs and node characteristics is proposed,which can dynamically merge nodes according to the similarity of node characteristics during the characteristic aggregation process to achieve parallel dimensionality reduction of large-scale power grid topology,node number and characteristics.Secondly,a generator-specified classifier model for the leading instable generators is proposed.Through global attention aggregation,the relative motion information of generators of the whole network is integrated into each generator characteristic vector,so that the identification model of leading instable generators can cope with the number of generator in structure and has good generalization ability.Finally,the model is verified in the actual large-scale power grid,and the effect and application performance of the key links are visually analyzed.