EKF-MCMC data assimilation framework for real-time state estimation and uncertainty quantification in reactor thermal-hydraulic analysis

The integration of data assimilation with small modular reactors offers a transformative opportunity to improve nuclear energy systems in safety, efficiency, and adaptability. By combining model predictions with observational data, data assimilation enhances state estimation accuracy and helps mitigate uncertainties in reactor safety analysis. However, its application is hindered by the complexity of two-phase flow models, which complicates intrusive assimilation methods, while non-intrusive alternatives rely heavily on extensive sampling or data-driven pretraining. To address these challenges, this study develops an extended Kalman filter-based data assimilation framework integrated with the RELAP5 two-fluid two-phase flow model. This approach enables robust real-time data fusion and efficient state-variable correction without the computational expense of sampling. Validation using LOFT L2-6 experimental data shows that the framework improves the accuracy of observed variables by 5%–60%, while also reliably estimating unobserved states. Furthermore, the development of an extended Kalman filter-Markov Chain Monte Carlo framework enables joint state-parameter estimation. This integrated approach not only corrects system states but also delivers parameter posterior estimates and inverse uncertainty quantification, thereby supporting model development, parameter calibration, safety margin assessment, and related applications. This work provides an efficient and reliable solution for data assimilation in reactor thermal-hydraulic analysis.