Abstract: Large language models (LLMs) and their derived multimodal large models have driven a new AI revolution due to their powerful generation and generalization capabilities. However, they have limitations such as hallucination problems and poor interpretability. Knowledge graphs (KGs) have the capabilities of explainable reasoning results and incremental knowledge updates, whereas, its interactive capabilities are weak. This paper reviews the development history, key technologies, advantages and limitations of KGs and LLMs. Focusing on the characteristics of power data and operational features, this paper analyzes mainstream approaches of applying KGs and LLMs in the power domain. A technical architecture that integrates KGs and LLMs for the power sector is established. The feasibility of each application scenario is analyzed in detail. Finally, the paper points out future challenges and potential research directions in this field.