Research on Intelligent Question-Answering Systems for Electrical Knowledge Based on Large Model Technology

The intelligent question-answering system for electrical knowledge can be used to popularize electrical knowledge and increase public understanding of the power industry. It can also serve as an auxiliary tool for training and evaluating the knowledge and skills of industry professionals, improving their expertise. Current electrical knowledge question-answering systems often rely on general models and lack specialized optimization for the electrical field, making it challenging for the system to generate more accurate and detailed answers based on domain-specific knowledge. This paper proposes PowerGPT, an intelligent question-answering system based on large language models and an electrical knowledge base, which provides knowledge enhancement for answer text generation. The system uses ChatGLM2-6B as the foundational model and fine-tunes it using LoRA. Additionally, the system employs a multi-expert model approach to subdivide the question-answering tasks. Comparative experiments show that PowerGPT outperforms question-answering systems lacking knowledge enhancement, model fine-tuning, and multi-expert models in the field of electrical knowledge. It demonstrates different answering strategies for various types of questions, mitigates the issue of model hallucination, and improves the accuracy and automation of electrical knowledge question-answering.