面向核电站操作员情境意识的动态贝叶斯与机器学习框架
摘要
操作员情境意识是复杂核电控制环境中人类可靠性的关键但难以捕捉的决定因素。现有的评估方法,如SAGAT和SART,仍停留在静态、回溯且脱离驱动操作风险的认知动态方面。为克服这些限制,本研究引入了情境意识的动态贝叶斯机器学习框架(DBML SA),这是一种统一的方法,将概率推理与数据驱动智能融合在一起,实现情境意识建模的量化、可解释和预测。该框架利用2007年至2021年的212个运行事件报告,重构了11个绩效影响因素(PSF)在多个认知层次上的因果时间结构。贝叶斯组件实现了不确定性下的情境意识可靠性的时间演化推断,而神经网络组件则建立了从PSF到SART评分的非线性预测映射,实现了平均绝对百分比误差为13.8%,并与主观评估在统计上保持一致(p > 0.05)。结果表明,训练质量和压力动态是情境意识退化的主要驱动因素。总体而言,DBML SA超越了传统的问卷式评估方法,通过实现实时认知监控、灵敏度分析和早期预警预测,为下一代数字主控制室的人机可靠性管理铺平了道路。
引用
@article{arxiv.2603.19298,
title = {A Dynamic Bayesian and Machine Learning Framework for Quantitative Evaluation and Prediction of Operator Situation Awareness in Nuclear Power Plants},
author = {Shuai Chen and Huiqiao Jia and Tao Qing and Li Zhang and Xingyu Xiao},
journal= {arXiv preprint arXiv:2603.19298},
year = {2026}
}
备注
This article is withdrawn due to a technical error identified after submission in the data processing and modeling workflow described in Sections 3 -- 4. The issue affects feature construction and statistical estimation, which may compromise the reliability of the reported results. The authors withdraw this version to avoid potential misunderstanding. A revised study may be submitted in the future