Abstract: To address the limitation of insufficient interaction ability of spatiotemporal interaction models in trajectory prediction, this study proposes a spatiotemporal interaction trajectory prediction model (GT-BiGRU) that integrates the graph attention mechanism (GAT) and temporal encoding. This proposed model comprises three modules: a feature enhancement module, a spatiotemporal interaction module, and a dynamic output module. The algorithm combines a bidirectional gated recurrent unit (BiGRU), which excels at time-series prediction, with a graph attention mechanism (GAT), which is adept at representing spatial relationships. Temporal encoding is incorporated to facilitate comprehensive interaction between these two components across both temporal and spatial dimensions. In the graph attention layer, the target features and environmental features from the input data are designated as target nodes and neighbor nodes, respectively, to construct a graph structure. At the same time, temporal-encoding vectors are embedded in this layer to enhance the GAT’s dynamic updating ability of node features, effectively capturing the periodic patterns and trending characteristics inherent in trajectory sequences. The adaptive mechanism of the graph attention layer and the temporal-encoding enhancement strategy are integrated with the bidirectional propagation mode of the BiGRU. This integration helps identify key node information within historical trajectories, thereby improving the fitting accuracy of the generated trajectories. The experimental results show that GT-BiGRU outperforms the comparison model across mean square error (MSE), mean absolute error (MAE), average displacement error (ADE), and final displacement error (FDE).