A multimodal trajectory prediction model was proposed for vehicle interaction behaviors in real traffic environments. Time series, spatial sequences, and social behavior were integrated for modeling and predicting future vehicle trajectories. Time-dependent features such as vehicle speed and acceleration were extracted from historical data using long short-term memory (LSTM). The interaction intensity between vehicles was quantified by graph convolutional networks (GCN) through a weighted adjacency matrix. A convolutional social pooling module was introduced to extract local interaction features and enhance hierarchical spatial interaction patterns. A multimodal prediction mechanism was adopted, where a multilayer perceptron (MLP) was used to compute the probability distribution of driving behaviors. The LSTM decoder was combined to generate future trajectory distributions corresponding to specific driving behaviors. Experimental results demonstrate that the proposed model achieves strong performance in vehicle trajectory prediction. Compared to baseline models, the root mean square error is reduced by 27.9% on average, and the negative log-likelihood is reduced by 56.45% on average.