Knowledge Retrieval Enhancement for Power System Based on Adaptive Dynamic Text Segmentation and Hierarchical Multi-Query

To address the issue of long document correlation information retrieval and understanding in complex power system data, this paper proposes a power system knowledge retrieval enhancement method based on dynamic text segmentation and hierarchical multiple queries. An adaptive dynamic text segmentation method is proposed, which dynamically adjusts the window size and stride according to the text content to adapt to the dynamic segmentation of text under different information densities. A hierarchical multiple query mechanism is proposed, which decomposes the original problem into multiple sub-problem queries, and improves the depth and breadth of retrieval by reasoning and integrating the answers to each subproblem, ensuring the comprehensiveness and accuracy of retrieval. Experiments are conducted on six datasets, and the results showed that the F1 scores under two vector databases (Chroma and Milvus) are 52.52 and 55.85 respectively, indicating that the proposed method can effectively enhance the power system knowledge retrieval and understanding capabilities, providing technical references for further optimization and prediction of power systems.