Due to the continuous promotion of China’s carbon peak and carbon neutrality strategies, the proportion of new energy installed capacity has increased from 29% in 2018 to 38% in 2023. The high-frequency response characteristics of power electronic devices reduce the overall inertia of the power grid by nearly 40%, and the propagation speed of power grid faults accelerates from seconds to milliseconds. For this change, traditional security detection systems mainly rely on threshold alarms and offline calculations, making it difficult to effectively track the dynamic changes in power grid topology and integrate heterogeneous data from multiple sources under real-time requirements. Therefore, it is gradually becoming difficult to meet the security monitoring needs of new power systems. Therefore, this article constructs an online security detection framework for power grids based on dynamic graph neural networks (DGNN). This framework can integrate the topology relationship of the power grid, steady-state data collected by SCADA system, and transient timing data provided by PMU together. The experimental results showed that the accuracy of dynamic graph neural network in fault detection reached 98.2%, with a false alarm rate of only 1.1%, and the single inference time was controlled within 32 milliseconds. All performance indicators are significantly better than other comparison methods, fully meeting the real-time requirements of online detection below 100 milliseconds. This study provides a technical method based on graph calculation for online safety monitoring of new power systems, which has important engineering application value for improving the resilience and reliability of power grid operation.
He et al. (Thu,) studied this question.