【摘要】 传统的人员可靠性分析（HRA）主要依赖于人工操作者参与的实验。然而，随着新一代核反应堆设计中界面与操作规程的创新，传统HRA数据逐渐失去了适用性。这些新系统面临着显著挑战：一是缺乏足够的实际运行经验，二是传统“人在回路”实验方法本身存在低效问题。本文提出了一种基于ACT-R认知架构的人类数字孪生体（HDTs）方法，全面实现了操作员认知过程的建模，包括声明性记忆、程序性知识以及感知—运动模块，用于预测操作员的任务完成时间与错误概率。实验在清华大学的高温气冷堆（HTGR）模拟器上开展，涵盖了多种工况，包括带界面切换的多参数查询、不切换界面的参数查询、单纯的三向交流操作，以及按钮交互操作。由于三向交流操作在流程中占主导地位，有效降低了错误发生率，从而增强了实验流程的可靠性。为了进一步评估ACT-R在人因错误预测方面的能力，设计了两项快速响应任务：一项为识别反应堆启停、正常运行、事故与停堆等不同状态；另一项为界面上报警信号的快速枚举。结果表明，基于ACT-R的人类数字孪生仿真框架不仅能够精准预测操作员的反应时间，还能有效估算操作错误概率，展示出其作为核设施主控室HRA数据采集替代方案的可行性。更多还原

【Abstract】 Traditional human reliability analysis has mainly relied on experiments involving real human operators. However, with the introduction of novel interfaces and operating procedures in the design of next-generation nuclear reactors, traditional HRA data has gradually lost its applicability. These new systems have faced significant challenges: Firstly, a lack of sufficient operational experience. Secondly, the inherent inefficiency exists with traditional “human-in-the-loop” experimental methods. This work has introduced, for the first time, a human digital twin approach based on the ACT-R cognitive architecture, which has fully enabled the modeling of operators’ cognitive processes, including declarative memory, procedural knowledge, and perceptual-motor modules, for predicting task completion time and error probability. Experiments have been conducted on the High Temperature Gas-Cooled Reactor(HTGR) simulator at Tsinghua University, covering multiple operational scenarios, including multi-parameter queries with interface switching, parameter queries without interface switching, standalone three-phase AC operations, and button-based interaction tasks. Since three-phase AC operations have dominated the procedural workflow and have effectively reduced the likelihood of errors, the overall reliability of the experimental procedure has been enhanced. To further evaluate ACT-R’s capability in predicting human error, two rapid-response tasks have been designed: one for recognizing different reactor states, such as startup, normal operation, accident, and shutdown; the other for rapid enumeration of alarm signals on the interface. The results have shown that the ACT-R-based human digital twin simulation framework has not only accurately predicted operator response times but has also effectively estimated operation error probabilities, demonstrating its feasibility as an alternative HRA data acquisition solution for main control room environments.更多还原