核能
计算机科学
核工程
应急响应
决策支持系统
功率(物理)
核电站
环境科学
核物理学
物理
数据挖掘
量子力学
医学
工程类
医疗急救
作者
Xingyu Xiao,Peng Chen,Ben Qi,Jingang Liang,Jiejuan Tong,Haitao Wang
标识
DOI:10.1016/j.pnucene.2025.105908
摘要
As climate change and other global challenges increase the likelihood of unforeseen emergencies, the limitations of human-driven strategies become more pronounced. While decision support systems do exist and are implemented by various nuclear power plant vendors, current approaches are largely tailored to predefined accident sequences. There remains a critical need for adaptive decision support frameworks that can handle unforeseen or unprevented emergency scenarios not fully covered by existing procedures or best practices. This study addresses the urgent need for agile decision-making in response to various unforeseen incidents through a novel approach, EvoScenTree (a Scenario-driven method with evolvable interactive agents using event trees for dynamic emergency maintenance decision support). This method is a hybrid approach that integrates both expert knowledge and data-driven techniques. From the perspective of expert knowledge, we leverage insights from event tree analysis, which consists of three key tasks: initiating sub-event analysis, event tree header event analysis, and decision recommendations. From a data-driven standpoint, we design LLM-driven agents: task executors responsible for carrying out critical emergency response actions, and task validators that evaluate the appropriateness and effectiveness of those actions. Through iterative interactions between these two types of agents, emergency measures are continuously updated. Finally, we use nuclear power plants as a demonstration of the proposed EvoScenTree. Our findings indicate that the designed agents are not only effective but also outperform existing approaches in processing previously unencountered incident scenarios. This paper demonstrates that EvoScenTree significantly enhances the rapid formulation of emergency maintenance decision-making.
科研通智能强力驱动
Strongly Powered by AbleSci AI