Welcome to the PSAM 18 Abstract Status page.
Abstract KR220Full Paper + Presentation
Uncertainty Characterization and Reduction in HRA for Advanced Nuclear Reactors: Bayesian Parameter Estimation Approach
Advanced nuclear reactors, such as small modular reactors and microreactors, present new challenges in terms of risk assessment and Human Reliability Analysis (HRA), due to their new design characteristics and operational contexts. One example of a distinct characteristic of these advanced reactors, compared to conventional light-water reactors, is increased automation that can significantly influence operators’ performance during emergency conditions. A systematic literature review of HRA applications (2020–2025) for microreactors revealed remaining research gaps: studies mostly relied on qualitative categorization, frequencies obtained from generic databases not specific to nuclear contexts, or simplified nominal human error probabilities (HEP) values without a formal cognitive basis.
The NRC developed the Integrated Human Event Analysis System (IDHEAS) as the state-of-the-art HRA method to support its risk-informed regulation. Its database, documented in IDHEAS-DATA, is primarily built from simulator data and other sources specific to conventional Light Water Reactors (LWRs). In general, the introduction of increased automation in advanced reactors is expected to shift cognitive failure modes and performance-influencing factors to contexts where existing HRA data points are relatively limited.
In relation to this challenge, this study addresses two research questions: (i) how much additional uncertainty can limited data introduce into HRA in an advanced reactor context, and (ii) how much additional data is needed to reduce it to acceptable levels? This research proposes a Bayesian inference approach to explicitly construct probability distributions that represent uncertainties in human error probabilities and performance-influencing factor (PIF) weights. Probabilistic programming is implemented using the PyMC Python package. This paper demonstrates sensitivity analyses, where synthetic data with the same structure as IDHEAS-DATA are generated with varying characteristics, such as the number of data points for each PIF state and an imbalanced distribution of data points across PIF states. These results provide generalizable insights into which data characteristics are most influential on HRA uncertainty and how HRA uncertainty can be reduced most effectively through additional data collection.
← Check another abstractThe NRC developed the Integrated Human Event Analysis System (IDHEAS) as the state-of-the-art HRA method to support its risk-informed regulation. Its database, documented in IDHEAS-DATA, is primarily built from simulator data and other sources specific to conventional Light Water Reactors (LWRs). In general, the introduction of increased automation in advanced reactors is expected to shift cognitive failure modes and performance-influencing factors to contexts where existing HRA data points are relatively limited.
In relation to this challenge, this study addresses two research questions: (i) how much additional uncertainty can limited data introduce into HRA in an advanced reactor context, and (ii) how much additional data is needed to reduce it to acceptable levels? This research proposes a Bayesian inference approach to explicitly construct probability distributions that represent uncertainties in human error probabilities and performance-influencing factor (PIF) weights. Probabilistic programming is implemented using the PyMC Python package. This paper demonstrates sensitivity analyses, where synthetic data with the same structure as IDHEAS-DATA are generated with varying characteristics, such as the number of data points for each PIF state and an imbalanced distribution of data points across PIF states. These results provide generalizable insights into which data characteristics are most influential on HRA uncertainty and how HRA uncertainty can be reduced most effectively through additional data collection.