Estimating the quantitative relation between PSFs and HEPs from full-scope simulator data

For probabilistically assessing the risks of nuclear power plants, human reliability analyses (HRAs) have been conducted to systematically predict human error probabilities (HEPs) of significant tasks that might affect system safety. To improve quality in the HRA results, solid empirical evidence for quantitative relations between performance shaping factors (PSFs) and HEPs is required. For generating the empirical evidence, the HRA data including human reliability data and contextual data should be collected and the quantitative relations between PSFs and HEPs should be properly estimated. In this study, in order to validate the statistical estimation approach for the relation between PSFs and HEPs, the simulation records of operator training programs that were performed via full-scope simulators were collected and analyzed by the HuREX (Human Reliability data EXtraction) framework. The simulator data including more than 10,000 data points are then statistically analyzed by a logistic regression analysis method. From the statistical process, the significant variables affecting human reliability are deduced, and the PSF multipliers for each significant variable are estimated. The potentials and challenges of the statistical approach are discussed from the obtained results.