Intelligent optimization and evaluation techniques for thermal-hydraulic safety analysis of nuclear power systems

Thermal-hydraulic safety analysis of the nuclear reactor system features complex models, strong multi-parameter coupling, and stiff, highly nonlinear, ill-posed equations. Key technical issues are: first, how to improve the accuracy of key models and enhance the fidelity of system analysis codes while ensuring solution stability; second, how to optimize and update calculation results based on real-time data in code calculation; third, how to efficiently and accurately determine critical parameter weights in result evaluation technology; fourth, how to reasonably analyze the reliability of component models and perform optimization analysis. This review focuses on accuracy improvement and intelligent evaluation in reactor thermal-hydraulic safety analysis. It reviews intelligent optimization and evaluation technologies for nuclear power system analysis and discusses their application throughout the lifecycle of system analysis code development, optimization, evaluation, and application. To improve the prediction accuracy of basic models for nuclear power system analysis, an optimization method for key models based on artificial intelligence algorithms is developed. To address the low prediction accuracy and narrow applicability of constitutive equations, auxiliary equations, and other closure models, deep learning is used to construct AI models with high-dimensional nonlinear mapping capabilities. A combination of experimental data-driven approaches and physics-informed coupling strategies is adopted to improve prediction accuracy and applicability while ensuring computational robustness. Considering the accumulation of prediction errors over time, short-term corrective prediction for key reactor parameters is introduced to provide decision-making support. Data assimilation is employed to achieve real-time prediction optimization in reactor system analysis; correction methods for field and model parameters are developed based on measured data to update real-time prediction results and calibrate theoretical models, thereby improving the accuracy of dynamic model states and enhancing real-time prediction capability. For intelligent evaluation, a multidimensional framework for result evaluation and uncertainty quantification is constructed. Targeting the difficulty of evaluating real-time multi-source, multidimensional, non-uniform time-step data and the need to capture characteristic points and trends under special reactor transients, dynamic intelligent evaluation technologies integrating data dimensionality reduction, multi-parameter evaluation, and sensitivity analysis are developed. Accounting for modeling errors introduced by assumptions, simplifications, and approximations, as well as random data errors from measurement methods, surrogate-model acceleration and intelligent sampling are used to quantify the impact of input parameter uncertainty on calculation results, thereby identifying important input parameters and supporting uncertainty reduction and improvement of system analysis code models.Finally, based on prediction and evaluation results, reliability assessment of key system components using intelligent algorithms and deep learning is carried out to support multi-objective optimization design. Considering the randomness of physical processes and uncertainties of models and numerical methods, digital validation tests and fault tree analysis are employed to assess the reliability of nuclear power system equipment and reduce the risk of failure caused by accumulated uncertainties. In view of the multidimensional performance requirements and strong constraint–feedback characteristics of nuclear reactor systems, intelligent optimization techniques and deep learning algorithms for high-dimensional discrete data are integrated to realize multi-objective optimization of system analysis methods, thus meeting intelligent design needs for complex systems with multi-parameter performance constraints. Future work will establish best practice guidelines for intelligent optimization and evaluation technologies in nuclear reactor thermal hydraulic analysis and develop novel AI-based thermal hydraulic safety analysis methods, providing efficient and reliable tools for optimization design and safety review.