Hybrid Retrieval for Power Equipment Standard Knowledge

This paper addresses the accuracy and efficiency challenges in knowledge retrieval for power industry technical standards. Based on the coexistence of documents and metadata in semantic webs, we propose a large-model optimization method using hybrid retrieval techniques. By combining keyword search with metadata-based search, the method supports both semantic retrieval and knowledge retrieval tasks, significantly improving recall and precision for complex technical terms and long-tail queries. Experiments conducted on a power technical standards corpus indicate that the model attains a Top-5 retrieval accuracy of 80.7%, with strong robustness to fuzzy queries and synonymous terms. Furthermore, by incorporating domain-adaptive pretraining and knowledge enhancement strategies, the model effectively understands power industry-specific standard clauses, diagrams, and cross-document relationships. This study provides a reusable technical framework for knowledge retrieval in specialized vertical domains and has been validated for practical engineering use in power enterprise standard management systems.