Abstract: Aiming at the problem that the current power equipment selection design support system is difficult to cope with large-scale and high-complexity selection scenarios, with slow response speed and low design accuracy, a vector retrieval-enhanced intelligent design support system based on large language models is designed. Firstly, a professional knowledge vector database is constructed to encode design knowledge under diverse specific engineering scenarios into high-dimensional vectors and stored in the vector database. Then, grouped query attention and dual-block attention mechanisms are used to optimize the multi-round retrieval process, and the fast retrieval of expertise is achieved by calculating similarities between input requirements and knowledge base vectors. Finally, the multi-round dialogue and context tracking under cue engineering are used to optimize the selection design schemes and enhance system intelligence. The experimental results demonstrate the system performs well in several complex power equipment selection scenarios and can meet the power industry's demand for efficient and intelligent selection design support.

Key words: power equipment selection, large language model, vector retrieval, natural language processing, intelligent design