人工智能技术在核电领域的应用

© 2025 Electric Power Research Institute, Inc. All rights reserved.w w w . e p r i . c o mRobert AustinEPRI16th China International Exhibition on Nuclear Power IndustryApril 29, 2025Riding the Wave© 2025 Electric Power Research Institute, Inc. All rights reserved.2Vision To be a world leader in advancing science and technology solutions for a clean energy futureMissionAdvancing safe, reliable, affordable, and clean energy for society through global collaboration, science and technology innovation, and applied research.Together…Shaping the Future of Energy® © 2025 Electric Power Research Institute, Inc. All rights reserved.3Enhancevalue to members and societyDevelop and evaluate solutions while engaging national labs, technical partners, and international stakeholdersA trusted source of informationIndustry, lawmakers, policymakers, thought leaders, regulators, financial communityGuiding climate and grid strategyEconomy-wide modeling to optimize sustainability, reliability, and resiliency solutionsPushing the technological frontierEPRI’s Role© 2025 Electric Power Research Institute, Inc. All rights reserved.4Overview§ AI is coming§ EPRI is there§ What now?© 2025 Electric Power Research Institute, Inc. All rights reserved.5© 2025 Electric Power Research Institute, Inc. All rights reserved.6Image generated with MS Copilot© 2025 Electric Power Research Institute, Inc. All rights reserved.7Performance Increases from Model Scaling(percentile on common exams compared to human test-takers)Uniform Bar ExamLSATGRE (Quantitative)AP Calculus BCAP Chemistry0 10 20 30 40 50 60 70 80 90 100 GPT-3.5 (2022)GPT-4 (2023)© 2025 Electric Power Research Institute, Inc. All rights reserved.8© 2025 Electric Power Research Institute, Inc. All rights reserved.9AI is NOT new in nuclear§ “Early” AI– Advanced Pattern Recognition– Computer vision– Early language models§ Collaboration opportunities– AI/DX Summits (2021, 2022, Jan 2025)– Data-Driven Decision Making (3DM)Learning how to ride the wave© 2025 Electric Power Research Institute, Inc. All rights reserved.10Large Language ModelRetrieval Augmented Generation (RAG): BasicsUser queryRetrieval from the database via vector similarity searchVectorized using a finetuned Embeddings ModelEmbedded query[most relevant text chunks retrieved as context][User query][System prompt]LLMGenerated answer + sourcesEPRI and other reports© 2025 Electric Power Research Institute, Inc. All rights reserved.11How to ride the AI wave§ “Don’t tell me what to do. Tell me how to do it.” Matt Hall, Evergy§ How to develop a strategy, project selection, business cases:– Digital Transformation Roadmap§ How to execute, testing, metrics - see– Automating Corrective Action Programs in the Nuclear Industry (3002023821)– Good Practice Guide for Setting Up Large Language Model-Powered Projects (3002030935)§ Data sharing– “No one should have to have their o