Dynamic probabilistic risk assessment (DPRA) is an advanced risk analysis method that more realistically captures the progression of events during accident scenarios that may affect the safety of a nuclear power plant (NPP). This method integrates the modeling of random events that may occur during an accident scenario with deterministic simulations of system behavior. This provides a detailed, accurate understanding of risk scenarios when modeling interdependencies between events, which are not captured in conventional PRA methods, and enables high-fidelity accident progression simulations. As a result, DPRA supports an improved understanding of system-level risks and the impact on plant operation as a whole.

A DPRA analysis typically involves simulating thousands to millions of accident scenarios, each of which produces high-dimensional time-series data with numerous variables and temporal dependencies. The scale and complexity of this data introduce significant challenges when trying to efficiently and effectively analyze the time-series data to extract actionable risk insights. These challenges create a critical barrier to the adoption of DPRA methods within the nuclear energy industry.

This work investigates the use of advanced machine learning techniques to enable efficient analysis of high-dimensional DPRA simulation data. Specifically, this work explores methods for dimensionality reduction of DPRA simulation results while preserving key information relevant to accident progression and risk. The proposed approach aims to facilitate improved identification of key system behaviors and accident patterns within large simulation datasets. By improving the interpretability of DPRA results, this work supports more effective, rapid risk assessment, which can promote a wider adoption of DPRA methods, ultimately contributing to more efficient, cost effective, and safer nuclear power plant operation.