As regional hub substations, the main busbar configurations of 500kV substations critically impact grid reliability and economic efficiency. To address this, this paper designs an intelligent design support system for substation main busbar schemes enhanced by vector retrieval based on large language models. The set-pair analysis method is introduced to assess the risks of main busbar configurations, which are then combined with quantitative economic indicators to form multidimensional data. Building upon this foundation, principal component analysis (PCA) is applied to comprehensively evaluate the multidimensional data indicators, ultimately determining the most optimal main busbar configuration. Experimental results demonstrate that the large language model, trained on a knowledge base of 500kV substation main busbar configurations, exhibits outstanding performance in long-text generation, achieving a maximum ROUGE-L score of 0.9876. Furthermore, the comprehensive evaluation method based on PCA further validates the effectiveness of the proposed design methodology. This research addresses the power industry’s demand for efficient, intelligent substation main busbar configuration design support.
