Transient Stability Prediction Method for Power Systems Based on Spatiotemporal Feature Fusion

To improve the accuracy and robustness of power system transient stability prediction, this paper proposes a deep learning evaluation method that integrates image modeling with temporal feature representation. The method employs Gramian Angular Field (GAF) to transform generator rotor angle sequences into two-dimensional images, enhancing the separability between stable and unstable states. A dual-channel architecture is constructed, combining a convolutional neural network (CNN) for extracting GAF-encoded spatiotemporal patterns and gated recurrent units (GRU) for modeling dynamic temporal evolution, with multi-head attention mechanisms introduced to achieve weighted fusion of cross-modal features for stability classification. Experiments were conducted on the IEEE 39-bus New England test system, generating a simulation dataset covering various fault types and operating conditions. Comparative analyses were performed against typical models including SVM, RF, and GRU. Results demonstrate that the proposed model outperforms baseline methods in key metrics such as accuracy and comprehensive scores, while exhibiting strong robustness under noise disturbances and missing data scenarios. The study validates the model’s advantages in capturing spatiotemporally coupled features during power system transients, demonstrating significant engineering applicability and generalization potential.