The safe and efficient operation of smart Intelligent vehicles relies heavily on accurate trajectory prediction techniques. Existing methods improve prediction accuracy by introducing scene context information, but lack the causal perspective to explain why scene context improves prediction performance. In addition, current multimodal trajectory prediction methods are mostly target-driven and implicitly fused, relying too much on the density of candidate targets, as well as ignoring the road rule constraints, which leads to a lack of anthropomorphic properties in the model's prediction results. To this end, this paper proposes a novel multimodal trajectory prediction model, CCTP-Net, which introduces causal interventions in the encoding phase to balance the influence of spatio-temporal features of the learned scene context on the trajectory. A node refinement strategy based on the cognitive properties of human drivers is designed between the feature aggregator and the decoder by selectively traversing the lane graph in order to identify key road node features. The extracted important nodes are finally used for multimodal trajectory decoding after counterfactual reasoning. CCTP-Net experiments on the publicly available dataset nuScenes show that the model has significant advantages in multimodal trajectory prediction in complex scenarios, verifying its effectiveness and reliability. This study provides new theoretical perspectives and technical paths for trajectory prediction of intelligent connected vehicles, which is expected to promote the further development of related technologies.