Identification of dominant instability modes in power systems based on spatial-temporal feature mining and TSOA optimization

Abstract

The recognition of the transient dominant instability mode is of great significance for rapidly and accurately formulating transient emergency decisions in power systems. In response to the challenge of accurately distinguishing between angle instability and voltage instability, which are coupled in actual power grids, this paper explores the mapping relationship between simulation data and the stable state of the system, as well as the dominant instability mode. The method enables real-time identification of the dominant instability mode, which bypasses complex physical mechanisms. Firstly, spatio-temporal feature mining is conducted, where convolutional neural networks are employed to learn crucial local features of transient curves, and bidirectional gated recurrent unit s utilized to learn transient features over time sequences. Next, a multihead attention mechanism is introduced to enhance sensitivity to important time steps in the sequence data. Finally, the transit search optimization algorithm optimizes the global model parameters, further increasing the accuracy of the model. Using the IEEE 10-machine and 39-node system as an example for simulation, the results validate that the proposed method exhibits significant advantages in terms of accuracy and applicability compared with other machine learning methods.