Assessing and improving model fitness in MOCABA data assimilation

Hoefer, Axel; Buss, Oliver

doi:10.1016/j.anucene.2021.108490

Cited by 5 publications

(1 citation statement)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Assumptions, such as an uncertainty being relative to the underlying measured quantity or a quantitity being non-negative, can be rigorously taken into account. The exact treatment of nonlinear relationships also enables the use of other probability distributions related to the (multivariate) normal distribution, such as the log-normal distribution for strictly positive variables, e.g., [16], and the application of transformations to obtain approximately normally distributed variables in cases where the original variables are not normally distributed, e.g., [27].…”

mentioning

confidence: 99%

Nuclear data evaluation with Bayesian networks

Schnabel¹,

Capote²,

Koning³

et al. 2021

Preprint

View full text Add to dashboard Cite

Bayesian networks are graphical models to represent the deterministic and probabilistic relationships between variables within the Bayesian framework. The knowledge of all variables can be updated using new information about some of the variables. The Bayesian Generalized Linear Least Squares method can be regarded as an inference method for Bayesian networks of variables with multivariate normal priors and linear relationships between them. We show that relying explicitly on the Bayesian network interpretation enables large scale inference and gives more flexibility in incorporating prior assumptions and constraints into the nuclear data evaluation process, such as the constraints that some cross sections equal linear combinations of other cross sections and that all cross sections must be non-negative. The latter constraint is accounted for by a nonlinear transformation and therefore we also discuss inference in Bayesian networks with non-linear relationships between variables. Using Bayesian networks, the evaluation process yields more detailed information, such as posterior estimates and uncertainties of all statistical and systematic errors associated with the experiments. We further elaborate on a sparse Gaussian process construction that can be well integrated into the Bayesian network framework and applied to, e.g., the modeling of energy-dependent model parameters, model deficiencies of the physics model or energy-dependent systematic errors of experiments. We present three proof-of-concept examples that emerged in the context of the neutron data standards project and in the ongoing international evaluation efforts of 56 Fe. In the first example we demonstrate the modelization and explicit estimation of relative energy-dependent error components associated with experimental datasets. Then we show that Bayesian networks in combination with the outlined Gaussian process construction may be applied to an evaluation of 56 Fe in the energy range between one and two MeV, where it is difficult to obtain satisfactory evaluations by R-Matrix and nuclear model fits. Finally, we present a model-based evaluation of 56 Fe between 5 MeV and 30 MeV with a consistent and statistically sound treatment of model deficiencies. The R scripts to reproduce the Bayesian network examples and the nucdataBaynet package for Bayesian network modeling and inference have been made publicly available.

show abstract

mentioning

confidence: 99%

Nuclear data evaluation with Bayesian networks

Schnabel¹,

Capote²,

Koning³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

Framework for the correct treatment of model input parameters for Bayesian updating problems in nuclear engineering

Jędrzejczyk,

Kopka,

Foad

2025

Annals of Nuclear Energy

View full text Add to dashboard Cite

Оценка Качества Решений В Задачах Ассимиляции Нейтронно-Физических Данных: Обзор Подходов И рекомендации По Их Применению

Андрей Алексеевич,

Ольга Николаевна

2024

Управление Большими Системами: Сборник Трудов

View full text Add to dashboard Cite

Приводится обзор современных подходов к диагностике и оценке качества результатов ассимиляции данных в задачах моделирования нейтронно-физических процессов в объектах использования атомной энергии. Несмотря на широкое применение методов ассимиляции нейтронно-физических данных в последнее время, вопросам диагностики и оценки качества, достоверности и надежности результатов ассимиляции не уделяется должного внимания. В работе систематизированы и описаны все известные метрики и подходы к оценке качества используемых ковариационных данных, показатели информативности и подобия реакторно-физических экспериментов друг другу и в отношении целевого объекта, способы выявления противоречивых экспериментов и диагностики качества решений на основе различных статистических показателей. Обсуждаются области применения различных метрик и подходов, их преимущества и недостатки, а также авторское видение места каждого из показателей и возможная последовательность их применения при реализации процедур ассимиляции в рассматриваемой предметной области. В работе также приводятся рекомендации, направленные на то, чтобы избежать нефизических решений, минимизировать компенсаторные эффекты при корректировке модельных параметров, включая нейтронные константы и технологические параметры, обеспечить возможность вовлечь в анализ эксперименты, носящие противоречивый характер. The article presents a comprehensive review of state-of-the-art approaches to diagnosing and assessing the quality of data assimilation results in neutronics modeling problems. Despite the widespread use of data assimilation procedures worldwide to refine the parameters of neutronics models based on reactor experiment results there is a lack of attention given to the issues of diagnostics and quality assessment in this specific area. This stage is crucial in ensuring the reliability and accuracy of assimilation results. By adhering to relevant recommendations, it is possible to avoid obtaining non-physical solutions, minimize compensatory effects when adjusting initial data, and include contradictory experiments in the analysis. The article discusses the most popular metrics and approaches for assessing the quality of covariance data, as well as indicators of informativeness and similarity between reactor physics experiments and the target object. It also covers methods for identifying contradictory experiments and diagnosing the quality of the solution using various statistical indicators. The article highlights the areas of application for different metrics and approaches, as well as their advantages and disadvantages, providing recommendations for their use.

show abstract

Assessing and improving model fitness in MOCABA data assimilation

Cited by 5 publications

References 23 publications

Nuclear data evaluation with Bayesian networks

Nuclear data evaluation with Bayesian networks

Framework for the correct treatment of model input parameters for Bayesian updating problems in nuclear engineering

Оценка Качества Решений В Задачах Ассимиляции Нейтронно-Физических Данных: Обзор Подходов И рекомендации По Их Применению

Contact Info

Product

Resources

About