Diagnosis Method for Pressurized Water Reactor LOCA Accidents Based on CART-LSTM Algorithm
-
摘要: 失水事故（LOCA）是压水堆的典型事故，事故可能诱发反应堆堆芯熔化，因此及时诊断LOCA非常重要。长短期记忆（LSTM）神经网络是一种改进的循环神经网络（RNN），能够更好地捕捉时序数据中的长期依赖关系，被广泛应用于与时序有关的故障诊断中。分类与回归树（CART）是一种常用的分类方法，具有分类速度快、准确率高、可读性强等特点。本文提出一种基于CART-LSTM的压水堆LOCA诊断方法，利用LOCA的数据集对诊断模型进行训练并优化参数，然后将训练好的模型用于LOCA诊断，从而实现对LOCA的早期快速诊断。结果表明，基于CART-LSTM的诊断方法能够准确判断LOCA的位置以及具体的破口尺寸。
-
关键词:
- 失水事故（LOCA） /
- 故障诊断 /
- 长短期记忆（LSTM） /
- 决策树
Abstract: Loss of Coolant Accident (LOCA) is a typical accident in pressurized water reactors, which may induce core melting. Therefore, timely diagnosis of LOCA is very important. Long Short-Term Memory (LSTM) neural network is an improved Recurrent Neural Network (RNN) that can better capture long-term dependencies in temporal data and is widely used in fault diagnosis related to temporal data. Classification and Regression Tree (CART) is a commonly used classification method, which has the characteristics of fast classification speed, high accuracy, and strong readability. Therefore, a pressurized water reactor LOCA diagnosis method based on CART-LSTM is proposed. The paper trains and optimizes the diagnostic model using the LOCA dataset, and then uses the trained model for LOCA diagnosis, thereby achieving early and rapid diagnosis of LOCA accidents. The results indicate that the diagnosis method based on CART-LSTM can accurately determine the position and specific size of LOCA accidents. -
图 2 LSTM细胞结构
${c_t}$、${h_t}$和$ \boldsymbol{x}_t $—t时刻的细胞状态、隐藏层状态和外部输入；${f_t}$、${i_t}$和${o_t}$—遗忘门、输入门和输出门的激活值；$ \tilde{c}_t $—状态候选值；$\sigma $—sigmoid激活函数；x—输入；tanh—激活函数。
Figure 2. Structure of LSTM Cells
图 6 LOCA位置诊断结果随时间的变化
Figure 6. The Variation of LOCA Position Diagnosis Results over Time
表 2 不同破口位置的破口尺寸 %
Table 2. Break Sizes at Different Break Positions
热管段 冷管段 稳压器波动管 故障标签 0.05 0.05 0.05 1 0.1 0.10 0.10 2 0.2 0.50 0.50 3 0.3 1.00 1.00 4 0.5 2.00 2.00 5 0.8 4.00 4.00 6 1.0 6.00 6.00 7 1.2 8.00 8.00 8 1.5 10.00 10.00 9 
下载: 导出CSV
表 3 不同时间窗口长度下LSTM的诊断效果
Table 3. Diagnosis Effect of LSTM under Different Time Window Lengths
时间窗口长度 准确率/% 1 97.21 3 98.10 5 98.52 7 99.59 9 99.57 11 99.64
下载: 导出CSV
表 4 CART-LSTM的LOCA诊断结果
Table 4. LOCA Diagnosis Results of CART-LSTM
诊断内容 评价指标 准确率/% 精准率/% 召回率/% 破口位置诊断 99.82 99.89 99.49 破口尺寸诊断 99.59 99.59 99.60
下载: 导出CSV
-
[1] ZHOU G, PENG M J, WANG H. Enhancing prediction accuracy for LOCA break sizes in nuclear power plants: a hybrid deep learning method with data augmentation and hyperparameter optimization[J]. Annals of Nuclear Energy, 2024, 196: 110208. doi: 10.1016/j.anucene.2023.110208 [2] 李仕鲜,刘井泉,沈永刚. 基于神经网络方法的LOCA事故诊断[J]. 核技术,2017, 40(8): 080604. [3] 赵云飞,张立国,童节娟,等. BP神经网络在AP1000核电站事故诊断应用中的初步研究[J]. 原子能科学技术,2014, 48(S1): 480-484. [4] 刘永阔,夏虹,谢春丽,等. BP-RBF神经网络在核电厂故障诊断中的应用[J]. 原子能科学技术,2008, 42(3): 193-199. doi: 10.7538/yzk.2008.42.03.0193 [5] 曹桦松,孙培伟. 基于PCA-RBF神经网络的小型压水堆故障诊断方法研究[J]. 仪器仪表用户,2021, 28(1): 49-55. doi: 10.3969/j.issn.1671-1041.2021.01.013 [6] 耿柯繁,许分钦,庞波,等. 基于LSTM的核反应堆一回路传感器故障智能检测[J]. 自动化仪表,2023, 44(S1): 366-370,375. [7] 邓志光,吴茜,朱加良,等. 基于改进LSTM的核电厂传感器故障诊断研究[J]. 自动化仪表,2023, 44(6): 115-120. [8] 慕昱,夏虹,刘永阔. 基于邻域粗糙集和决策树算法的核电厂故障诊断方法[J]. 原子能科学技术,2011, 45(1): 44-47. doi: 10.7538/yzk.2011.45.01.0044 [9] WANG Y Q, SUN P W. Kernel principle component analysis and random under sampling boost based fault diagnosis method and its application to a pressurized water reactor[J]. Nuclear Engineering and Design, 2023, 406: 112258. doi: 10.1016/j.nucengdes.2023.112258 [10] SUN Z G, WANG G T, LI P F, et al. An improved random forest based on the classification accuracy and correlation measurement of decision trees[J]. Expert Systems with Application, 2024, 237: 121549. doi: 10.1016/j.eswa.2023.121549 [11] 孙运淼,林锋,周激流. 长短时记忆网络在移动场景中的应用研究进展[J]. 现代计算机,2017(35): 10-15. [12] DA SILVA D G, DE MOURA MENESES A A. Comparing Long Short-Term Memory (LSTM) and bidirectional LSTM deep neural networks for power consumption prediction[J]. Energy Reports, 2023, 10: 3315-3334. doi: 10.1016/j.egyr.2023.09.175 [13] ZUBAIR M, ABABNEH A, ISHAG A. Station black out concurrent with PORV failure using a Generic Pressurized Water Reactor simulator[J]. Annals of Nuclear Energy, 2017, 110: 1081-1090. doi: 10.1016/j.anucene.2017.08.023 -
下载:
计量
- 文章访问数: 15
- HTML全文浏览量: 6
- PDF下载量: 0
- 被引次数: 0
