Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Zheng, X.; 玉置 等史; 塩津 弘之; 杉山 智之; 丸山 結
Proceedings of Asian Symposium on Risk Assessment and Management 2017 (ASRAM 2017) (USB Flash Drive), 11 Pages, 2017/11
Nuclear reactor severe accident simulation involves uncertainties, which may result from incompleteness of modeling of accident scenarios, selection of alternative models and unrealistic setting of parameters during the numerical simulation, etc. Both deterministic and probabilistic methods are required to reach reasonable estimation of risk for severe accidents. Computational codes are widely used for the deterministic accident simulations. Bayesian approaches, including both parametric and nonparametric, are applied to the simulation-based severe accident researches at Japan Atomic Energy Agency (JAEA). In the paper, an overview of these research activities is introduced: (1) Dirichlet process models, a nonparametric Bayesian approach, are applied to source term uncertainty and sensitivity analyses; (2) Gaussian process models are applied to the optimization for operations of severe accident countermeasures; (3) Nonparametric models, include models based on Dirichlet process and K-nearest neighbors algorithm, are built to predict the chemical forms of fission products. Simplified models are integrated into the integral severe accident code, THALES2/KICHE; (4) We have also launched the research of dynamic probabilistic risk assessment (DPRA), and because a great number of accident scenarios will be generated during DPRA, Bayesian approaches would be useful for the boosting of computational efficiency.
Zheng, X.; 石川 淳; 杉山 智之; 丸山 結
Nuclear Engineering and Technology, 49(2), p.434 - 441, 2017/03
被引用回数：4 パーセンタイル：36.71(Nuclear Science & Technology)Containment venting is one of essential measures to protect the integrity of the final barrier of a nuclear reactor during severe accidents, by which the uncontrollable release of fission products can be avoided. The authors seek to develop an optimization approach, from a simulation-based perspective, to the venting operations by using an integrated severe accident code, THALES2/KICHE. The effectiveness of containment venting strategies needs to be verified via numerical simulations based on various settings of venting conditions. The number of iterations, however, needs to be controlled for cumbersome computational burden of integrated codes. Bayesian optimization is an efficient global optimization approach. By using Gaussian process regression, a surrogate model of the "black-box" code is constructed. It can be updated simultaneously whenever new simulation results are acquired. With predictions via the surrogate model, upcoming locations of the most probable optimum can be revealed. The sampling procedure is adaptive. The number of code queries is largely reduced for the optimum finding, compared with pure random searches. One typical severe accident scenario of a boiling water reactor is chosen as an example. The research demonstrates the applicability of the Bayesian optimization approach to the design and establishment of containment-venting strategies during severe accidents.
Zheng, X.; 石川 淳; 杉山 智之; 丸山 結
Proceedings of 13th Probabilistic Safety Assessment and Management Conference (PSAM-13) (USB Flash Drive), 10 Pages, 2016/10
Containment venting is one of essential measures to protect the integrity of the final barrier of a nuclear reactor, by which the uncontrollable release of fission products can be avoided. The authors seek to develop an optimization approach to the planning of containment-venting operations by using THALES2/KICHE. Factors that control the activation of the venting system, for example, containment pressure, amount of fission products within the containment and pH value in the suppression chamber water pool, will affect radiological consequences. The effectiveness of containment venting strategies needs to be confirmed through numerical simulations. The number of iterations, however, needs to be controlled for cumbersome computational burden of severe accident codes. Bayesian optimization is a computationally efficient global optimization approach to find desired solutions. With the use of Gaussian process regression, a surrogate model of the "black-box" code is constructed. According to the predictions through the surrogate model, the upcoming location of the most probable optimum can be revealed. The number of code queries is largely reduced for the optimum finding, compared with simpler methods such as pure random search. The research demonstrates the applicability of the Bayesian optimization approach to the design and establishment of containment-venting strategies under BWR severe accident conditions.
