Research on Rapid Analysis Method of Ex-core Detector Response Based on BP Neural-network Algorithm
-
摘要: 堆外探测器响应描述了中子注量率与电流信号之间的对应关系，对反应堆的安全运行有着至关重要的作用。针对确定论方法和蒙特卡罗方法均无法兼顾堆外探测器响应计算效率与计算精度这一问题，采用基于反向传播（BP）神经网络的算法完成堆外探测器响应的快速计算；基于堆芯核设计系统CMS对我国现有百万千瓦级压水堆堆芯进行物理建模，堆芯内燃料组件排布方式和燃耗变化作为BP神经网络的输入，相对应的燃料组件排布方式以及不同燃耗下的堆外探测器响应作为BP神经网络的输出，构建了3层BP神经网络模型并进行了优化；经过计算验证，优化后模型能够较为快速地计算堆外探测器响应，且预测值与堆芯核设计系统CMS计算值相比误差较小，有较好的工程应用前景，为计算堆外探测器响应提供了新思路。
-
关键词:
- 反应堆 /
- 堆外探测器响应 /
- 反向传播（BP）神经网络 /
- 快速分析
Abstract: The response of the ex-core detector describes the corresponding relationship between neutron fluence rate and current signal, which plays a crucial role in the safe operation of the reactor. In response to the problem that both deterministic and Monte Carlo methods cannot balance computational efficiency and accuracy in calculating the response of ex-core detectors, a back propagation (BP) neural network algorithm is used to quickly calculate the response of ex-core detectors. Based on the core design system CMS, the physical modeling of the existing million-kilowatt pressurized water reactor core in China was carried out. The fuel assembly arrangement and burnup changes in the core are used as the inputs of the BP neural network, and the corresponding fuel assembly arrangement and detector responses outside the reactor under different burnup are used as the outputs of the BP neural network. A three-layer BP neural network model is constructed and optimized. After calculation verification, the optimized model can quickly calculate the response of the ex-core detector, and the predicted value has a smaller error compared to the core design system CMS calculation value. It has good engineering application prospects, providing a new idea for calculating the response of the ex-core detector.-
Key words:
- Reactor /
- Ex-core detector response /
- BP neural-network /
- Rapid analysis
-
图 6 样本集数量变化对平均绝对误差的影响
Figure 6. The Impact of Changes in Sample Set Quantity on Average Absolute Error
表 1 燃料组件种类
Table 1. Fuel Assembly Type
燃料组件类型 燃料组件个数 燃料富集度/% 是否含有可燃毒物 A 4 1.8 否 B 33 2.4 是，含12根可燃毒物棒 C 12 2.4 是，含16根可燃毒物棒 D 36 3.1 否 E 8 3.1 是，含12根可燃毒物棒 F 8 3.1 是，含16根可燃毒物棒 H 56 3.7 否 
下载: 导出CSV
表 2 四分之一堆芯归一化功率分布
Table 2. Normalized Power Distribution of One-forth Core
组件 a b c d e f g h 1 0.619 0.831 1.033 1.313 1.424 0.638 0.618 0.237 2 0.801 0.755 1.056 1.535 1.045 0.799 0.800 0.248 3 1.086 1.069 1.108 1.014 1.601 1.443 0.659 4 1.359 1.556 1.021 1.011 1.332 1.378 0.399 5 1.432 1.049 1.619 1.377 1.497 0.934 6 0.624 0.797 1.450 1.398 0.943 7 0.613 0.792 0.645 0.401 8 0.230 0.244
下载: 导出CSV
表 3 燃料组件对探测器响应的归一化贡献
Table 3. Normalized Contribution of Fuel Assemblies to Detector Response
组件 a b c d e f g h 1 0 0 0 0 0.002 0.01 0.046 0.165 2 0 0 0 0.002 0.013 0.073 0.401 1.823 3 0 0 0.002 0.013 0.079 0.485 3.006 4 0 0.002 0.013 0.080 0.492 3.030 19.421 5 0.002 0.013 0.079 0.492 3.088 19.331 6 0.010 0.073 0.485 3.030 19.331 7 0.046 0.401 3.006 19.421 8 0.165 1.823
下载: 导出CSV
表 4 不同激活函数对本问题MSE的影响
Table 4. The Influence of Different Activation Functions on the MSE in this Question
函数名称 Relu Elu Tanh Softmax MSE 1.6676 1.7917 2.1997 无法收敛
下载: 导出CSV
表 5 BP神经网络的参数
Table 5. Network Structure of BP Neural-network
参数 类型或数值 学习率 0.1 优化器 Adam 丢弃率 0 迭代次数 750
下载: 导出CSV
表 6 不同计算方法计算堆外探测器响应所耗时间
Table 6. Time Required to Calculate Ex-core Detector Response using Different Calculation Methods
计算方法 计算时间 MCNP 168 h CMS 40 min BP神经网络 577 ms
下载: 导出CSV
-
[1] SERRA P L S, MASOTTI P H F, ROCHA M S, et al. Two-phase flow void fraction estimation based on bubble image segmentation using randomized Hough transform with neural network (RHTN)[J]. Progress in Nuclear Energy, 2020, 118: 103133. doi: 10.1016/j.pnucene.2019.103133 [2] 徐龙飞,沈华韵,魏军侠,等. 去射线效应堆外探测器响应函数计算研究[J]. 原子能科学技术,2020, 54(8): 1460-1467. doi: 10.7538/yzk.2019.youxian.0521 [3] 丁谦学,梅其良. 堆外探测器响应函数三维空间分布计算[J]. 强激光与粒子束,2017, 29(3): 036013. doi: 10.11884/HPLPB201729.160191 [4] 丁谦学,夏春梅,梅其良. CAP1000核电厂堆外探测器响应函数计算方法研究[J]. 核科学与工程,2016, 36(2): 257-262. doi: 10.3969/j.issn.0258-0918.2016.02.018 [5] 林海鹏,李国栋,陈法国,等. BP神经网络算法预测多组分材料中子屏蔽效果方法研究[J]. 辐射防护,2020, 40(6): 516-521. [6] 于志翔,邹树梁,徐守龙,等. 基于BP神经网络的船用反应堆屏蔽设计快速计算功能研究[J]. 核电子学与探测技术,2016, 36(2): 209-213. doi: 10.3969/j.issn.0258-0934.2016.02.022 [7] 王端,王威策,潘翠杰,等. 基于自适应BP神经网络的压水堆堆芯换料关键参数的预测方法[J]. 原子能科学技术,2020, 54(1): 112-118. doi: 10.7538/yzk.2019.youxian.0016 [8] 周旭华,李富,韩松,等. 堆外探测器读数与堆内功率分布的关系研究[J]. 核电子学与探测技术,2010, 30(2): 166-170. doi: 10.3969/j.issn.0258-0934.2010.02.004 [9] ZHOU L W, GARG D, QIU Y, et al. Machine learning algorithms to predict flow condensation heat transfer coefficient in mini/micro-channel utilizing universal data[J]. International Journal of Heat and Mass Transfer, 2020, 162: 120351. doi: 10.1016/j.ijheatmasstransfer.2020.120351 [10] 韦子豪,王端,王东东,等. 神经网络-遗传复合算法在压水堆堆芯换料设计中的应用[J]. 原子能科学技术,2020, 54(5): 825-834. doi: 10.7538/yzk.2019.youxian.0788 -
计量
- 文章访问数: 22
- HTML全文浏览量: 4
- PDF下载量: 4
- 被引次数: 0
