To address the issues where the existing knowledge graph completion models based on graph neural networks fail to distinguish the contributions of different neighboring entities to the central entity, and only rely on the extraction of structural information while neglecting the important semantic information carried by the inherent types of entities, leading to insufficient embedding representation capabilities and limiting the prediction performance of the model, this paper proposed a knowledge graph completion model based on dynamic neighborhood aggregation and type enhancement. Firstly, the model employs an attention mechanism to dynamically adjust entity neighborhood aggregation, refining the contributions of neighboring entities to the central entity to obtain higher quality entity embeddings and enhances the semantic representation of the entity embeddings using entity type information. Secondly, a type graph is constructed to capture the semantic associations of entity types reflected in the nature of relationships, improving the quality of relationship embeddings. Finally, the enhanced entity and relationship embeddings are input into the decoder, and a scoring function is used to score the input feature triples to complete the knowledge graph completion task. Comparing the performance of the proposed method with the baseline method CompGCN on the FB15k-237 and NELL-995 datasets, the MRR and the hits@3 evaluation indicators achieved significant improvements of 1.9, 2.3 and 2.1, 0.9 percentage points, respectively. Experimental results indicate that the model can effectively utilize entity type information to enhance both entity and relationship embeddings, thereby improving the model's accuracy in prediction tasks.