<p>To address the issue of insufficient modeling of dynamic interaction behaviors between vehicles in existing vehicle trajectory prediction methods, this paper proposes a vehicle trajectory prediction method (STDGAT) based on a dynamic graph neural network that integrates spatiotemporal features. First, the core of the method lies in constructing a high-precision dual spatiotemporal graph structure and defining novel matrix weight coefficients to simulate real driving environments. Spatial interaction features are then extracted using the dynamic graph attention network (DGAT). Second, the temporal convolutional network (TCN) is employed to extract temporal features from the target vehicle’s historical trajectory. Subsequently, an adaptive gating unit is used to fuse the temporal and spatial features. These fused features are then embedded into an encoder–decoder structure incorporating an attention mechanism. The encoder generates spatiotemporal feature vectors, which are further refined by the attention mechanism to capture critical spatiotemporal features. Finally, the decoder takes these features as input and performs multi-step iterative decoding to predict the future trajectory of the target vehicle. Given the depth of the model and the scale and dynamic nature of the spatiotemporal graph data processed, this study leverages a high-performance computing environment. To validate the effectiveness of the proposed STDGAT method, iterative training is conducted on the open-source NGSIM dataset. Comparative experiments are performed against trajectory prediction models including GRIP++, HiVT,ADAPT , and GSTCN. The results demonstrate that, compared to other mainstream models, the proposed model achieves superior prediction accuracy in both highway and urban scenarios.</p>