This paper proposes a Retrieval-Augmented Generation (RAG) framework that combines knowledge graphs and vector databases to enhance the performance of large language models (LLMs). This method first retrieves relevant subgraphs from the knowledge graph and then integrates this information with that from the vector database to enhance the generative capabilities of the model. Through experimental evaluations in the medical field, the results show that this framework significantly improves the quality and accuracy of responses generated by the model when handling complex questions, outperforming RAG frameworks that use only a single knowledge base. This approach effectively leverages the strengths of different knowledge bases, optimizing information retrieval.