GPU-accelerated approaches for multi-unit level 3 PSA: Focusing on dispersion, evacuation, and dose assessment

This study proposes a Graphics Processing Unit (GPU)-accelerated computational framework aimed at addressing the significant computational challenges inherent in key modules of Level 3 Probabilistic Safety Assessment (Level 3 PSA) for multi-unit nuclear power plant accidents. Specifically, it focuses on three of the most computationally intensive processes—atmospheric dispersion modeling, agent-based evacuation dynamics, and radiological dose assessment. By integrating these modules into a unified, scalable simulation environment, the proposed GPU-based framework can simultaneously and efficiently handle numerous meteorological datasets, evacuee groups, nuclear power plant units, and Gaussian puffs, significantly enhancing computational performance compared to existing methodologies. Comparative performance evaluations demonstrate that this approach achieves speedups of over 1400 times relative to traditional Central Processing Unit (CPU)-based methods, effectively supporting near-real-time emergency response and precise risk assessments. Although additional aspects of Level 3 PSA—such as ingestion exposure pathways and economic impact assessments—are important, they are not within the scope of this study and can be addressed separately as post-processing or through dedicated modules. Instead, this study focuses on demonstrating the necessity and effectiveness of GPU acceleration in accurately capturing the spatiotemporal complexity and radiological release characteristics of multi-unit nuclear accidents.