军事知识服务平台
基于LoRA微调大模型的电力生产专家系统研究与实践
Research and Practice of Power Production Expert System based on LoRA Fine-tuning Large Model
摘要 本文针对当前水电站人工智能(AI)大模型应用存在的信息孤岛、数据解析困难、模型适配不足等瓶颈问题,提出了基于低阶适应(LoRA)微调大模型构建电力生产专家系统的方法并完成系统建设.通过整合电厂技术文档构建训练集,运用 LoRA技术对Qwen3-32B模型进行领域适配,结合多种工具和平台开发出集专家咨询、智能搜索与方案助手等功能于一体的电力专用智能体.实践表明,该系统显著提升了问答精确度、关联度和响应速度,在用户满意度、响应时间和业务流程优化等方面带来了可观的经济效益,为电力生产领域的智能化发展提供了有效解决方案和实践参考.
Abstract:
The paper addresses the bottleneck issues in the application of AI large models in hydropower stations,such as information silos,difficulties in data analysis and insufficient model adaptation.It proposes and implements a power production expert system based on the fine-tuning of large models using LoRA.By integrating technical documents from power plants to construct a training set,and employing LoRA technology for domain adaptation of the Qwen3-32B model,it develops a dedicated intelligent agent for power that encompasses expert consultation,intelligent search,and solution assistance functionalities.Practical results demonstrate that this system significantly improves the accuracy,relevance and response speed of Q&A interactions.It has brought considerable economic benefits in terms of user satisfaction,response time and optimisation of business processes,providing effective solutions and practical references for the intelligent development in the field of power production.
刊 名 大电机技术 ISTIC
年,卷(期): 2026, (z1)
分类号: TM612
机标分类号: TP393.09 F270.7 F426.61
在线出版日期: 2026年3月11日