Objective To address the limitation in current smart grid false data injection attack FDIA detection methods which utilize only spatial data features for attack identification while neglecting temporal correlations across sequential system states this paper proposes SBIGRU-HA a detection model integrating stacked bidirectional gated recurrent unit SBIGRU and hybrid attention HA . Methods First the SBIGRU extracted the system?? s temporal features within a given time period and captured the temporal relationships among the data. Meanwhile a residual network was introduced to fuse the original input with the temporal features captured by the SBIGRU. On this basis three attention mechanisms namely coordinate attention CA convolutional block attention module CBAM and parameter-free attention SimAM were integrated. These mechanisms were used to extract the spatiotemporal features of the data and assign higher weights to the features with injected attacks. Finally the obtained feature representation was fed into a linear layer and a Sigmoid layer to complete the attack detection. Results Simulation experiments were conducted on the IEEE-14 and IEEE-57 node test systems. The results demonstrated that the SBIGRU-HA model achieved detection accuracies of 98. 68% and 96. 36% and F scores of 98. 32% and 95. 39% respectively on these two test systems. Conclusion Compared with LSTM and GRU the SBIGRU-HA model demonstrates improvements across all detection metrics and is capable of identifying the specific locations of false data.