Speaker
摘要
在人机交互密集的安全关键系统中，人机界面（HMI）失效仍是导致人为错误的重要来源。然而，传统的人类可靠性分析（HRA）方法在很大程度上依赖定性判断，难以刻画界面设计因素对人为失误的定量影响。尽管近年来基于认知建模的方法引入了一定的量化机制，但其通常未能在动态运行情境中，将客观界面设计属性与人为错误概率建立直接关联。针对这一关键问题，本文提出了一种结构化的界面数据驱动概率建模框架——InSight-R，用于定量刻画界面设计属性与界面诱发人为错误之间的关联机制。
该框架融合了客观界面特征分析、实证行为数据以及层次化概率推断方法，实现了对界面诱发风险的路径级诊断。模型基于理论驱动的方向性关系，刻画了界面指标与操作者认知—行为状态之间的耦合机制，其中目标显著性、语义干扰和交互跨度等关键因素共同影响执行类错误与时间偏差类错误的发生概率。通过在高保真核电仿真平台上开展典型面板巡检任务实验，并结合持证操纵员与学生被试数据对模型进行验证，结果表明：视觉复杂性与语义复杂性的提升显著增加了执行错误和时间偏差的发生概率。
通过建立界面设计属性与人为错误概率之间的定量映射关系，InSight-R为人机界面相关绩效塑形因子的建模提供了数据驱动的细化路径，并为动态可解释的人类可靠性分析提供了重要支撑。
Abstract
Human–machine interface (HMI) failures remain a major contributor to human errors
in safety-critical systems, yet conventional human reliability analysis (HRA) largely
relies on qualitative judgment. Although recent cognitive modeling approaches introduce quantitative mechanisms, they rarely link objective interface design attributes
directly to human error probabilities in dynamic operational contexts. To address this
gap, we propose InSight-R, a structured interface data-driven probabilistic framework
that quantitatively models the association between interface design attributes and
interface-induced human errors. By integrating objective interface analytics, empirical
behavioral data, and hierarchical probabilistic inference, the framework enables pathlevel diagnosis of interface-induced risks. The model captures theoretically motivated
directional relationships between interface metrics and operator cognitive–behavioral
states, where target salience, semantic interference, and interaction span jointly predict
execution-related and time-deviation errors. The framework was evaluated on a highfidelity nuclear simulator using representative panel-inspection tasks with licensed
operators and student participants. Results show that increased visual and semantic
complexity is associated with higher probabilities of execution errors and temporal
deviations. By establishing a quantitative link between interface design attributes and
human error probabilities, InSight-R provides a data-driven refinement of HMI-related
performance shaping factors and supports dynamically interpretable human reliability
analysis.
| 关键词 | 人类可靠性分析，人机界面，界面诱发的人因错误，层次化概率建模，动态风险分析 |
|---|---|
| Keywords | Human reliability analysis, Human–machine interface, Interface-induced human error, Hierarchical probabilistic modeling, Dynamic risk analysis |
