基于大模型和区块链的电力知识问答系统设计与实现

摘 要：如今电力已成为我们日常生活中不可缺少的一部分。然而，人们对电力知识的了解还有待进一步提高，电力知识推广的供给形式有待进一步创新。在AI大模型赋能各行各业高速发展的大背景下，电力领域的大模型存在例如数量不足和用户隐私数据保护力度不够等诸多问题。本研究以智谱AI和清华大学KEG实验室联合发布的对话预训练模型ChatGLM3为基础，设计开发了一个具备回答电力专业知识能力的智慧问答系统。在客户端和服务端数据传输与存储的过程中，我们采用了区块链加解密算法对数据进行处理和包装，以提高数据安全性。首先，收集高质量数据集，结合ChatGPT等基于Transformer的大模型，增强ChatGLM3的泛化能力；其次，通过Lora微调和全参微调等方法优化模型，优化后模型准确率提升31.78%，模型幻觉率降低了33.5%；最后，我们在系统终端数据存储过程中采用加解密算法对用户数据进行加密，以确保用户数据的安全性。

Design and implementation of a power knowledge system based on large model and block-chain technology

Abstract: Nowadays electricity has become an indispensable part of our daily life. However, people''s understanding of electric power knowledge needs to be further improved, and the supply form of electric power knowledge popularization needs to be further innovated. In the context of the rapid development of AI large models enabling all walks of life, there are many problems such as insufficient number of large models in the power field and insufficient protection of user privacy data. Based on ChatGLM3, a dialogue pre-training model jointly released by Zhipu AI and KEG Lab of Tsinghua University, an intelligent question answering system with the ability to answer professional knowledge of electric power is designed and developed in this study. In the process of client and server data transmission and storage, we use blockchain encryption and decryption algorithms to process and package data to improve data security. First, high-quality data sets are collected and combined with large Transformer based models such as ChatGPT to enhance the generalization capability of ChatGLM3. Secondly, the model was optimized by Lora fine-tuning and full-parameter fine-tuning. After optimization, the model accuracy was increased by 31.78%, and the model hallucination rate was reduced by 33.5%. Finally, we use encryption and decryption algorithm to encrypt user data in the storage process of system terminal data to ensure the security of user data.