The "dual carbon" goals place higher demands on the high-quality development of the energy storage sector, necessitating the thorough exploration and utilization of knowledge in related fields, as well as the provision of precise knowledge recommendation services for specific scenarios. This study proposes a knowledge recommendation solution for the energy storage domain that integrates large language models and knowledge graphs. By considering both the availability and authority of data sources for energy storage patents, the study categorizes entities in the energy storage domain into eight types based on two dimensions: metadata and text granularity. Various machine learning models are compared experimentally, and the optimal model is selected for entity extraction. A knowledge graph for the energy storage domain is constructed based on entity-relation templates to generate triples. Three response strategies are developed: general Q A, patent recommendation, and knowledge combination recommendation. Leveraging the powerful semantic understanding and generation capabilities of large language models, the system automatically retrieves relevant information from the knowledge graph to fulfill precise domain knowledge recommendation tasks. Empirical results from the superconducting energy storage field demonstrate that the proposed method outperforms traditional LLM-KB methods in four types of tasks: single-target recommendation, dual-target recommendation, knowledge combination recommendation, and hybrid recommendation. The proposed method improves ROUGE-L and BLEU-4 scores by 20.47% and 16.57%, respectively, compared to traditional LLM-KB methods. This study provides new insights into precise knowledge recommendations for the energy storage domain by integrating knowledge graphs and large language models, enhancing the efficiency and effectiveness of user access to high-quality knowledge.