Study on Risk-informed Safety Evaluation and Optimization under Adaptive Sampling in Nuclear Power Plant
-
摘要: 为了满足先进反应堆安全性与经济性双重需求，本文将确定论安全分析与概率安全分析相耦合，以“风险指引的系统分析（RISA）”为基础，重点突破高精度要求下的抽样次数海量、计算效率低的难题，提出了一种自适应采样策略下的RISA优化方法，现实评估核电厂安全裕度。采用支持向量机（SVM）、K近邻（KNN）算法训练得到代理模型，替代大量的最佳估算加不确定性分析程序模拟；并且用自适应采样策略识别极限面，降低抽样范围和次数。以小破口失水事故（SBLOCA）为例进行测试验证，结果表明：相比于随机抽样RISA结果，代理模型预测的燃料包壳峰值温度（PCT）结果接近、计算时间减少50%以上。本方法可支撑RISA实际工程应用，为风险指引的设计、运行与管理提供现实精准的决策支持。
-
关键词:
- 核安全分析 /
- 自适应采样 /
- 代理模型 /
- 风险指引的系统分析（RISA） /
- 风险指引的安全裕度特性分析（RISMC）
Abstract: In order to meet the dual requirements of safety and economy of advanced reactors, the Deterministic Safety Analysis and Probabilistic Safety Analysis are coupled in this study. Based on the Risk-informed System Analysis (RISA), a RISA optimization method under adaptive sampling strategy is proposed to realistically evaluate the safety margin of nuclear power plants. It focuses on solving the problems of massive sampling times and low calculation efficiency under high precision requirements. Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) Algorithm are used to train the surrogate model, which replaces a large number of Best Estimation and Uncertainty Analysis code simulations. The adaptive sampling strategy is used to identify the limit surface and reduce the sampling range and times. Taking the Small Break Loss of Coolant Accident (SBLOCA) as an example, the test results show that, compared with the random sampling RISA results, the peak temperature of fuel cladding predicted by the surrogate model is close, and the calculation time is reduced by more than 50%. Therefore, this method can support the practical engineering application of RISA in the future, and provide realistic and accurate decision support for the risk-informed design, operation and management. -
图 4 自适应采样和蒙特卡罗抽样样本下的代理模型评价指标结果对比
Figure 4. Comparison of Evaluation Index Results of Surrogate Model under Adaptive sampling and Monte Carlo Sampling
图 7 PCT概率密度函数拟合曲线（贝叶斯估计）
Figure 7. PCT Probability Density Function Curve (Bayesian estimation)
表 1 关键不确定性参数及其范围
Table 1. Key Uncertainty Parameters and Their Ranges
参数名 抽样范围 分布 破口面积/cm2 5~10 均匀分布 高压安注流量/(kg·s−1) 12.5±1.25 均匀分布 堆芯燃料间隙热流密度/(W·m−2) 4.36±0.05 均匀分布 高压安注泵失效时间/s 失效率λ=1.2×10−5/h 指数分布 
下载: 导出CSV
表 2 自适应采样样本下的代理模型性能（650个样本）
Table 2. Surrogate Model Performance under Adaptive Sampling (650 samples)
参数名称 初始数据 第一次迭代 第二次迭代 第三次迭代 第四次迭代 K 3 4 1 2 2 R2 0.9048 0.8394 0.9038 0.8963 0.9047 EVS 0.9049 0.8400 0.9059 0.8967 0.9061 MAPE 0.0821 0.1057 0.0619 0.0795 0.0640
下载: 导出CSV
表 3 蒙特卡罗随机采样样本下的代理模型性能（800个样本）
Table 3. Surrogate Model Performance under Monte Carlo Random Sampling (800 samples)
参数名称 初始数据 第一次迭代 第二次迭代 第三次迭代 第四次迭代 K 13 14 18 21 9 R2 0.7969 0.8361 0.8458 0.8838 0.8129 EVS 0.7984 0.8370 0.8459 0.8841 0.8129 MAPE 0.0816 0.1195 0.0983 0.1010 0.0958
下载: 导出CSV
表 4 不同模型的RISA计算效率对比
Table 4. Comparison of RISA Calculation Efficiency of Different Models
模型 单次最长耗时/s 计算次数 总时间 RELAP系统仿真模型 1777 200 13 h 自适应采样的代理模型 6.67×10−5 30000 2 s
下载: 导出CSV
表 5 PCT分布特性结果对比
Table 5. Comparison of Computing Results of PCT Distribution
方法 拟合方法 PCT分布期望/K 95%分位数预测/K 样本的PCT最大值/K PCT后果条件超限概率 1 RELAP5仿真结果拟合 843.34 1654.04 1963.91 0.1599 2 代理模型预测 929.44 1597.03 1893.76 0.1304 3 蒙特卡罗大量抽样 821.28 1649.62 2109.26 0.1231
下载: 导出CSV
-
[1] DE VASCONCELOS V, SOARES W A, DA COSTA A C L, et al. Deterministic and probabilistic safety analyses[M]//RAM M, DAVIM J P. Advances in System Reliability Engineering. London: Academic Press, 2019: 43-75. [2] BEDFORD T, COOKE R. Probabilistic risk analysis: foundations and methods[M]. Cambridge: Cambridge University Press, 2001: 99. [3] U. S. Department of Energy. Light water reactor sustainability research and development program plan fiscal year 2009-2013: INL/MIS-08-14918[R]. Idaho Falls: Idaho National Laboratory, 2009. [4] HESS S M, DINH N, GAERTNER J P, et al. Risk-informed safety margin characterization[C]//Proceedings of the 17th International Conference on Nuclear Engineering. Brussels: ASME, 2009. [5] SMITH C, RABITI C, SZILARD R. Risk-informed safety margins characterization (RISMC) pathway technical program plan: INL/EXT-17-43243-Rev000[R]. Idaho Falls: Idaho National Lab. , 2017. [6] Szilard R H, Zhang H, Choi Y J. Risk-Informed Systems Analysis (RISA) Pathway Technical Program Plan[R]. Idaho National Lab.(INL), Idaho Falls, ID (United States), 2018. [7] 王照,李琼哲,郭建兵. 基于RISMC分析方法的全厂断电叠加汽动辅助给水失效事故研究[J]. 核动力工程,2023, 44(6): 186-192. doi: 10.13832/j.jnpe.2023.06.0186. [8] ALDEMIR T. A survey of dynamic methodologies for probabilistic safety assessment of nuclear power plants[J]. Annals of Nuclear Energy, 2013, 52: 113-124. doi: 10.1016/j.anucene.2012.08.001 [9] HOFER E, KLOOS M, KRZYKACZ-HAUSMANN B, et al. Dynamic event trees for probabilistic safety analysis[R]. Berlin: GRS, 2004. [10] HAKOBYAN A, ALDEMIR T, DENNING R, et al. Dynamic generation of accident progression event trees[J]. Nuclear Engineering and Design, 2008, 238(12): 3457-3467. doi: 10.1016/j.nucengdes.2008.08.005 [11] Risk-informed Methods and Applications in Nuclear and Energy Engineering: Modeling, Experimentation, and Validation[M]. Academic Press, 2023. 43-54. [12] ABDEL-KHALIK H S, BANG Y, KENNEDY C, et al. Reduced order modeling for nonlinear multi-component models[J]. International Journal for Uncertainty Quantification, 2012, 2(4): 341-361. doi: 10.1615/Int.J.UncertaintyQuantification.2012003523 [13] DI MAIO F, BANDINI A, ZIO E, et al. An approach based on Support Vector Machines and a K-D Tree search algorithm for identification of the failure domain and safest operating conditions in nuclear systems[J]. Progress in Nuclear Energy, 2016, 88: 297-309. doi: 10.1016/j.pnucene.2016.01.017 [14] BAE J, PARK J W, LEE S J. Limit surface/states searching algorithm with a deep neural network and Monte Carlo dropout for nuclear power plant safety assessment[J]. Applied Soft Computing, 2022, 124: 109007. doi: 10.1016/j.asoc.2022.109007 [15] DIAW A, MCKERNS M, SAGERT I, et al. Efficient learning of accurate surrogates for simulations of complex systems[J]. Nature Machine Intelligence, 2024, 6(5): 568-577. doi: 10.1038/s42256-024-00839-1 [16] MANDELLI D, SMITH C. Adaptive sampling using support vector machines: INL/CON-12-26354[R]. Idaho Fall: Idaho National Laboratory, 2012. [17] CHEN H Y, WANG H, WANG L C, et al. Evaluation of probabilistic safety margin of nuclear power plant based on optimized adaptive sampling method[J]. Annals of Nuclear Energy, 2024, 201: 110409. doi: 10.1016/j.anucene.2024.110409 [18] FUHG J N, FAU A, NACKENHORST U. State-of-the-art and comparative review of adaptive sampling methods for Kriging[J]. Archives of Computational Methods in Engineering, 2021, 28(4): 2689-2747. doi: 10.1007/s11831-020-09474-6 -
计量
- 文章访问数: 41
- HTML全文浏览量: 8
- PDF下载量: 9
- 被引次数: 0
