Physics-informed optimization for emergency radiation assessment with temporal correction under meteorological uncertainty

Abstract

Timely and accurate radiation dose assessment is essential for effective emergency response in nuclear accidents. However, meteorological uncertainties, especially in wind data, can lead to substantial discrepancies between simulated and observed plume behaviors, compromising situational awareness and decision-making. This study proposed a physics-informed optimization framework that integrates a physical radiation assessment model with a genetic algorithm to dynamically correct time-series wind field data and mitigate discrepancies caused by meteorological uncertainty. The physical model couples the Lagrangian puff model with the point kernel integration method. To improve efficiency, a dimensionality reduction approach simplifies the three-dimensional gamma dose integration to one dimension. The proposed framework was validated using the first venting scenario of Unit 1 at the Fukushima Daiichi Nuclear Power Plant. The temporal optimization significantly enhanced the alignment of estimated and observed plume passage times. Quantitatively, the optimization respectively reduces the fractional bias (FB) and the normalized mean square error (NMSE) at the Main Gate by 57.82 % and 90.69 %, while the improvements at MP8 station reached 97.88 % (FB) and 92.19 % (NMSE). The FAC2 (Fraction of predictions within a factor of two) at the Main Gate increased substantially from 9.5 % to 52.4 % post-optimization. These improvements demonstrate the effectiveness of the proposed method in enhancing predictive accuracy for emergency radiation dose assessment and optimizing operational decision-making under complex atmospheric conditions.