CPC H02J 3/24 (2013.01) [G06N 3/10 (2013.01); G06Q 50/06 (2013.01); G06N 3/045 (2023.01); G06N 3/08 (2013.01); H02J 2203/20 (2020.01)] | 7 Claims |
1. A method for monitoring short-term voltage stability of a power system, comprising:
obtaining a topology and post-fault time series of a current power system;
inputting the topology and the time series of the current power system into a trained spatial-temporal graph network model, the trained spatial-temporal graph network model being obtained by classification learning on a spatial-temporal graph network model based on a simulation sample dataset;
outputting a status of the short-term voltage stability of the power system; and
sending an alarm signal in response to outputting an unstable status of the short-term voltage stability of the power system;
wherein before obtaining the topology and the time series of the current power system, the method further comprises:
constructing the simulation sample dataset;
constructing the spatial-temporal graph network model; and
training the spatial-temporal graph network model based on the simulation sample dataset;
wherein constructing the spatial-temporal graph network model comprises:
constructing a spatial-temporal information incorporation module for extracting spatial-temporal characteristics, the spatial-temporal information incorporation module being formed by stacking a plurality of spatial-temporal information incorporation blocks, wherein a graph convolutional layer of the spatial-temporal information incorporation block is configured to extract spatial information, and a one-dimensional temporal convolutional layer of the spatial-temporal information incorporation block is configured to extract temporal information;
constructing a node layer block for weighting and summing data of a plurality of dimensions obtained by each node based on the spatial-temporal characteristics to obtain a node representation corresponding to each node; and
constructing a system layer block for standardizing the node representation corresponding to each node, taking an absolute value of the node representation, multiplying the absolute value with a system layer parameter processed by a softmax function, and outputting an assessment result by the softmax function.
|