Abstract

The safe operation of a nuclear power plant (NPP) can be guaranteed through the team effort of operators in the main control room (MCR). Among the various features, peer checks, concurrent verification, independent verification, and communication reconfirmation are major contributors to effective operations in the MCR. In the digital MCR environment of advanced NPPs, there are potential emerging issues of concern related to these contributors resulting from the use of PC-soft controls for reactor operations. The objective of this study is to investigate the development of quantitative indicators for estimating the implicit intentions of reactor operators as a way to mitigate such concerns. The proposed quantitative indicators support peer checks and concurrent/independent verifications for diagnosing and preventing human errors through communication enhancement in a digital technology-based MCR. A machine learning–based algorithm was used to classify two implicit intentions of agreement and disagreement. The classification was based on electroencephalography data measured from human subjects while they performed mock operational tasks using soft controls. The mock operational tasks were based on using a Windows-based nuclear plant performance analyzer (Win-NPA). Statistical analysis was performed on the measured data to identify significant differences between the agreement and disagreement judgments by the operators. An average classification accuracy of 72% was achieved by using a support vector machine classifier for the Win-NPA task with a low number of features across the various Brodmann areas. The methodology proposed in this study may also serve to enhance communications in conventional MCRs for human error minimization.

摘要

核电站（NPP）的安全运行可以通过主控制室（MCR）操作员的团队合作来保证。在各种功能中，对等检查、并发验证、独立验证和通信重新确认是 MCR 中有效操作的主要贡献者。在先进核电厂的数字 MCR 环境中，由于对反应堆运行使用 PC 软控制，因此存在与这些贡献者相关的潜在新问题。本研究的目的是调查用于估计反应堆运营商的隐含意图的定量指标的发展，以此作为减轻此类担忧的一种方式。提议的定量指标支持同行检查和并发/独立验证，以通过基于数字技术的 MCR 中的通信增强来诊断和预防人为错误。使用基于机器学习的算法对同意和不同意这两种隐含意图进行分类。该分类基于人类受试者在使用软控制执行模拟操作任务时测量的脑电图数据。模拟操作任务基于使用基于 Windows 的核电站性能分析器 (Win-NPA)。对实测数据进行统计分析，找出操作者判断一致与不一致的显着差异。通过对 Win-NPA 任务使用支持向量机分类器实现了 72% 的平均分类准确率，在各个 Brodmann 区域中具有少量特征。本研究中提出的方法还可用于增强传统 MCR 中的通信，以最大限度地减少人为错误。