Vector Retrieval-Based Intelligent Q&A System for Offshore Power Transmission and Substation Projects

Current systems for designing transmission and substations struggle with complex marine environmental impacts and enhancing efficiency for cross-sea lines, requiring intelligent, automated solutions. This paper presents an intelligent Q&A system for cross-sea transmission engineering, focusing on anti-corrosion material selection, construction methods, and maintenance strategies. The system combines vector retrieval and natural language generation to provide rapid and accurate access to design information and generate personalized solutions. It constructs a knowledge vector database and employs Grouped Query Attention (GQA) and Dual-Block Attention (DCA) mechanisms for optimized knowledge retrieval. Additionally, it uses prompt engineering and multi-round dialogue mechanisms to enhance design schemes and system intelligence. Experimental results show that the system significantly improves design efficiency and response speed, meeting the industry's need for efficient and intelligent design support.