Data-Driven Reduced-Order Models of Control Room Procedural Execution

Digitization of nuclear power plant operating procedures is a critical step toward enabling automated operations, intelligent operator support systems, and next-generation digital control rooms. However, transforming legacy paper-based procedures into machine-readable operational knowledge requires both structured procedural representations and methods for analyzing execution behavior across diverse plant conditions. This study presents a framework for extracting and analyzing procedural execution data from digitized emergency operating procedures using knowledge graphs, scenario-driven execution traces, and probabilistic risk assessment–based cut sets. Execution trajectories generated from operational scenarios are encoded as binary procedural visitation vectors and analyzed using singular value decomposition and latent semantic analysis (LSA) to identify dominant execution modes and semantically significant procedural patterns. The results show that, despite the combinatorial complexity of emergency operating procedures, including branching logic, conditional actions, response-not-obtained paths, and interprocedure transfers, the underlying execution behavior exhibits a low-dimensional structure governed by a limited set of plant-state variables. The dominant singular modes correspond to physically interpretable operational behaviors, including common response pathways, recovery strategies, and equipment availability dependencies. In addition, LSA identifies repeated and semantically related procedural actions that may inform human reliability analysis and assessments of the cognitive workload of operators. These findings demonstrate the potential of using AI-enabled procedural digitization and reduced-order modeling approaches to support automated operations, digital twins, and advanced decision-support systems for nuclear power plant modernization.