<p>FDIA maliciously tampers with state estimation results, posing a significant threat to the stable and secure operation of power cyber-physical systems. To address the issues of fixed topology dependence in spatial feature extraction and insufficient temporal feature extraction, a parallel dual-branch detection method integrating a high-order Chebyshev graph convolutional network and a GAF-Swin Transformer is proposed. Specifically, high-order Chebyshev polynomials are employed to dynamically characterize the topological structure of power grids, while a spatial self-attention mechanism is introduced to break the constraint of fixed adjacency matrices in traditional graph convolutions. The Gramian angular field is utilized to encode time-series measurement data into polar coordinate images, which preserves the absolute phase information in the time dimension. By integrating with an improved Swin Transformer module, multiscale temporal dependency features are extracted via a hierarchical local–global attention mechanism. A cross-modal feature fusion layer is designed to weight the contributions of spatial and temporal features, thereby achieving complementary information enhancement. Experiments were conducted by substituting actual data from a specific region of the Northwest China Power Grid into the IEEE 14, 39, and 118 bus systems. The results demonstrate that the proposed method outperforms other FDIA detection methods in terms of both detection accuracy and robustness.</p>