To improve the accuracy of vehicle trajectory prediction in complex traffic environments, a vehicle trajectory prediction model integrating spatial interaction features, temporal dynamic features, and driving intention information is proposed. First, an interaction graph was constructed based on the relative positions and motion states of the target vehicle and surrounding vehicles, and spatial interaction features were extracted by a graph attention mechanism. Second, Transformer was employed to capture long-range global dependencies, and LSTM was introduced to characterize local dynamic variations, thereby obtaining more comprehensive temporal features. Furthermore, a spatial self-attention mechanism and a GRU-LSTM structure were used to recognize lane-changing intentions, and the probability distributions of left lane change, right lane change, and lane keeping were output to characterize the uncertainty of driving behavior. Finally, the spatiotemporal features and driving intention probabilities were concatenated at the feature level into a comprehensive representation vector, and the future trajectory of the target vehicle was generated through a fully connected layer. The proposed model was validated on the NGSIM dataset and demonstrates excellent performance. Compared with S-LSTM, the RMSE of the proposed model is reduced by 13.8%, 45.0%, 41.2%, 40.5%, and 38.8% at prediction horizons from 1 s to 5 s, respectively. Compared with CS-LSTM, the RMSE is reduced by 8.2%, 15.8%, 39.7%, 37.5%, and 36.3%, respectively. The results show that the proposed model significantly improves the accuracy of trajectory prediction and provides reliable decision support for intelligent driving systems.