Abstract: Accurately predicting the motion trajectories of surrounding vehicles is crucial for the safety of autonomous vehicles. Realizing the trajectory prediction of multi-target vehicles depends not only on the historical trajectories but also requires clarifying the dynamic spatial interactions between vehicles and the temporal relationships between trajectories of different time series. However, existing trajectory prediction methods do not adequately consider the coupling effects of spatial interactions and temporal relationships, resulting in insufficient accuracy for multi-target vehicles trajectory prediction in highly interactive scenarios. This paper proposes a planning-coupled multi-target vehicles trajectory prediction network (PCTP-Net) that contains encoding, feature fusion, and trajectory decoding modules for modeling coupled interactions based on time and space. Firstly, the encoding module employs a bidirectional long short-term memory (Bi-LSTM) network to encode historical and planning trajectories of different time series, which combines the planning information of the ego vehicle with the prediction process of multi-target vehicles to realize the coupled interaction modeling based on time and space. Secondly, the feature fusion module introduces a convolutional social pooling layer to analyze the impact of trajectories with different temporal features on the prediction and captures the dynamic spatial interactions between vehicles. Finally, the trajectory decoding module proposes a trajectory prediction decoder that incorporates driving behavior decisions to improve the trajectory prediction accuracy of multi-target vehicles under interaction. The experiments on the NGSIM dataset and different traffic scenarios show that the proposed method can achieve accurate trajectory prediction in traffic scenarios with dense and highly interactive vehicles.