邢程亮,宋飞,王宇航,张留碗

• 1:清华大学求真书院
• 2:清华大学物理系

摘要(Abstract)：

随着人工智能技术的发展与应用，AI赋能物理实验的潜力日益凸显。然而由于物理实验创新性强、方法迭代快，因此传统通用大语言模型（LLM）在针对特定实验方法进行推理时需要将本地知识库全部注入，进而耗费大量上下文资源，使用成本较高。本研究提出一种融合动态上下文感知管理与协同推理架构的物理实验模型框架。该框架可以直接适配现有大语言模型，通过动态筛选与自动化闭环校验，提取高价值相关信息注入模型上下文，在显著降低推理成本、提升推理效率的同时，实现了与传统全量上下文注入方法相当的推理质量。该框架支持快速二次开发与多元化实验场景与仪器适配，有效解决了计算资源与成本瓶颈，为物理实验自动化与智能化提供了高效灵活、值得借鉴的创新范式。

关键词(KeyWords)： 物理实验;深度学习;上下文感知;检索增强生成;协同推理

Abstract:

Keywords:

基金项目(Foundation): 通专教育背景下大学实验物理教学内容的重构探索与实践（教育部拔尖计划2.0研究课题20211007）;; 清华大学AI赋能物理实验教学综合改革与立体化资源建设（53100100826）;清华大学自主科研计划资助（大学生学术研究推进计划20267020002）

作者(Author): 邢程亮,宋飞,王宇航,张留碗

DOI: 10.27024/j.wlygc.2026.06.25.02

参考文献(References)：

• [1] LIU N F,LIN K,HEWITT J,et al.Lost in the middle:How language models use long contexts.Transactions of the Association for Computational Linguistics,2023,12:157-173.
• [2] LIU J,LIN J,LIU Y.Tackling the inherent difficulty of noise filtering in RAG.arXiv,https://arxiv.org/abs/2601.01896,2026.
• [3] BROWN T B,MANN B,RYDER N,et al.Language Models are Few-Shot Learners.In Advances in Neural Information Processing Systems,2020,33:1877-1901.
• [4] 邢程亮，宋飞，张留碗.稳态法测量导热系数实验的温度补偿优化[J].大学物理，2026,45(1):105-110.XING C L,SONG F,ZHANG L W.The temperature compensation optimization for steady-state method thermal conductivity measurement experiments[J].College Physics,2026,45(1):105-110.(in Chinese)
• [5] LING C,ZHAO X,LU J,et al.Domain Specialization as the Key to Make Large Language Models Disruptive:A Comprehensive Survey.ACM Computing Surveys,2025,58(3):1-39.
• [6] MIN S,LYU X,HOLTZMAN A,et al.Rethinking the Role of Demonstrations:What Makes In-Context Learning Work?arXiv,https://arxiv.org/abs/2202.12837.
• [7] MA Y,et al.A Survey of Retrieval-Augmented Generation (RAG) for Large Language Models.2025 International Conferenceon Trustworthy Big Data and Artificial Intelligence (ICTBAI),2025:7-13.
• [8] GAO Y F,XIONG Y,GAO X Y,et al.Retrieval-Augmented Generation for Large Language Models:A Survey.arXiv,https://arxiv.org/abs/2312.10997.
• [9] 程军，王玉杰，张永春，等.弦振动形成驻波的理论分析和实验研究[J].大学物理实验，2025,38(3):41-44.CHENG J,WANG Y J,ZHANG Y C,et al.Theoretical analyses and experimental study of standing waves born from string vibration[J].Physical Experiment of College,2025,38(3):41-44.(in Chinese)
• [10] 车宇，丁望峰.稳态法测量导热系数实验的温度补偿[J].物理实验，2020,40(5):11-14.CHE Y,DING W F.Temperature compensation in the experiment of measuring thermal conductivity by steady state method[J].Physics Experimentation,2020,40(5):11-14.(in Chinese)
• [11] 潘永杲，周逸，闵琪涛，等.稳态法导热系数测定仪校准方法及其不确定度评定[J].计量与测试技术，2022,49(9):107-110.PAN Y G,ZHOU Y,MIN Q T,et al.Calibration method and evaluation of the uncertainty for steady thermal conductivity tester[J].Metrology & Measurement Technique,2022,49(9):107-110.(in Chinese)
• [12] KASNECI E,SESSLER K,K??CHEMANN S,et al.ChatGPT for good?On opportunities and challenges of large language models for education.Learning and Individual Differences,2023,103:102274.
• [13] C??RDOVA-ESPARZA D M.AI-powered educational agents:Opportunities,innovations,and ethical challenges[J].Information,2025,16(6):469.
• [14] QI J,XU Z,SHEN Y,et al.The art of SOCRATIC QUESTIONING:Recursive thinking with large language models.Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing,2023:4177-4199.
• [15] GHOSH A,BHATIA A.Generative AI in Personalized Learning:A Systematic Review of Adaptive Educational Technologies and Their Impact on Learning Outcomes.In:Joshi,A.,Ragel,R.G.,S.,K.(eds) ICT:Applications and Social Interfaces.ICTCS 2025.Lecture Notes in Networks and Systems,2026,vol 1872.Springer,Cham.
• [16] CHENG M,LUO Y,OUYANG J,et al.A survey on knowledge-oriented retrieval-augmented generation[J].arXiv,https://arxiv.org/abs/2503.10677.