Research on Diagnosis Method of Operational Events of Nuclear Reactor Based on Convolutional Long Short-term Memory Network and Artificial Whale Algorithm
-
摘要: 当核电厂发生异常后应及时诊断原因，以避免对运行人员和周围环境造成严重后果。本文利用卷积神经网络（CNN）和长短期记忆（LSTM）网络可更好地提取数据的局部特征和记忆时间序列信息的特征，研究基于卷积长短期记忆（CLSTM）网络和人工鲸鱼算法的核反应堆运行事件诊断技术。通过核电厂反应堆模拟机仿真实验对本文所述方法进行测试，最终测试准确率为99.91%，证明了本文所述研究方法的有效性。相关研究成果可作为核电厂运行事件的一种诊断方法，有利于提高运行事件诊断的智能化和信息化水平，为核电厂的少人值守甚至无人值守提供技术基础，提高公众对核电厂的认识与信赖。
-
关键词:
- 核反应堆 /
- 运行事件诊断 /
- 卷积长短期记忆（CLSTM） /
- 人工鲸鱼算法
Abstract: In case of abnormality in the nuclear power plant, the cause shall be diagnosed in time to avoid serious consequences to the operators and the surrounding environment. In this paper, convolutional neural network (CNN) and long short-term memory (LSTM) network can better extract the local characteristics of data and the characteristics of memory time series information, and study the operational event diagnosis technology of nuclear reactor based on convolutional long short-term memory (CLSTM) network and artificial whale algorithm. The method described in this paper was tested by the simulation experiment of nuclear power plant reactor simulator, and the final test accuracy is 99.91%, which proves the effectiveness of the research method described in this paper. The relevant research results can be used as a diagnosis method of nuclear power plant operational events, which is conducive to improving the intelligence and information level of operational event diagnosis, providing a technical basis for few or no people on duty in nuclear power plants, and improving the public's understanding and trust in nuclear power plants. -
图 1 基于CLSTM网络进行运行事件诊断流程图
Figure 1. Flow Chart of Operational Event Diagnosis Based on CLSTM Network
图 2 运行事件诊断算法的基本流程图
Figure 2. Basic Flow Chart of Operational Event Diagnosis Algorithm
图 5 人工鲸鱼群的全局最优解和平均适应度
Figure 5. Global Optimal Solution and Average Fitness of Artificial Whale Optimization
图 6 CLSTM最优结构的训练和测试准确率
Figure 6. Training and Testing Accuracy of CLSTM Optimal Structure
图 7 CLSTM最优结构训练与测试的交叉熵损失
Figure 7. Cross Entropy Loss of CLSTM Optimal Structure Training and Testing
表 1 稳态运行核电厂设计值与模拟机仿真值的对比
Table 1. Comparison Table of Steady-State Operating Nuclear Power Plant Design Values and Simulator Simulation Values
参数 设计值 仿真值 误差/% 稳压器压力/MPa 15.40 15.44 0.23 稳压器水位/m 7.598 7.59 0.12 稳压器波动管温度/℃ 345.09 344.86 0.06 主冷却剂流量/(kg·s−1) 4923.06 4923.97 0.02 压力容器出口温度/℃ 326.70 327.01 0.10 压力容器入口温度/℃ 292.58 293.98 0.10 压力容器水位/m 12.02 12.01 0.08 SG给水流量/(kg·s−1) 532.72 530.21 0.47 SG蒸汽产量/(kg·s−1) 526.30 526.53 0.04 SG出口蒸汽压力/MPa 6.70 6.72 0.30 SG出口蒸汽温度/℃ 285.8 285.6 0.07 SG—蒸汽发生器 
下载: 导出CSV
表 2 CLSTM的最优网络结构
Table 2. Optimal Network Structure of CLSTM
第1层卷积 卷积核的数量 34 卷积核的尺寸 2 Dropout的比例 0.148 第2层卷积 卷积核的数量 203 卷积核的尺寸 5 Dropout的比例 0.367 第3层卷积 卷积核的数量 234 卷积核的尺寸 5 Dropout的比例 0.495 第4层卷积 卷积核的数量 85 卷积核的尺寸 2 Dropout的比例 0.1 第1层CLSTM网络 LSTM网络堆叠数量 187 Dropout的比例 0.436 第2层CLSTM网络 LSTM网络堆叠数量 102 Dropout的比例 0.275 第3层CLSTM网络 LSTM网络堆叠数量 105 Dropout的比例 0.354
下载: 导出CSV
表 3 不同运行事件诊断模型的准确率和技术指标
Table 3. Accuracy and Technical Indexes of Different Operational Event Diagnosis Models
结合人工鲸鱼算
法的诊断模型FCN CNN GRU LSTM CLSTM 训练准确率 0.7643 0.9412 0.9808 0.9886 0.9946 测试准确率 0.8075 0.9626 0.9969 0.9987 0.9991 训练损失误差 0.5086 0.1738 0.0656 0.0398 0.0226 测试损失误差 0.3935 0.1403 0.0148 0.0084 0.0028
下载: 导出CSV
表 4 不同优化算法下的诊断准确率
Table 4. Diagnosis Accuracy under Different Optimization Algorithms
参数寻优算法 最佳准确率/% 平均准确率/% 遗传算法 94.86 94.22 基本粒子群算法 93.91 93.68 增强粒子群算法 95.71 95.12 人工鲸鱼算法 98.86 96.67
下载: 导出CSV
-
[1] SINGH P, SINGH L K. Design of safety critical and control systems of nuclear power plants using petri nets[J]. Nuclear Engineering and Technology, 2019, 51(5): 1289-1296. doi: 10.1016/j.net.2019.02.014 [2] ČEPIN M. Evaluation of the power system reliability if a nuclear power plant is replaced with wind power plants[J]. Reliability Engineering & System Safety, 2019, 185: 455-464. [3] HINES J W, DAVIS E. Lessons learned from the U. S. nuclear power plant on-line monitoring programs[J]. Progress in Nuclear Energy, 2005, 46(3-4): 176-189. doi: 10.1016/j.pnucene.2005.03.003 [4] YU L, NAZIR B, WANG Y L. Intelligent power monitoring of building equipment based on Internet of Things technology[J]. Computer Communications, 2020, 157: 76-84. doi: 10.1016/j.comcom.2020.04.016 [5] 温志斌,刘永阔,段智勇,等. PCA-SDG在反应堆冷却剂系统典型故障诊断中的应用[J]. 应用科技,2016, 43(5): 82-87. [6] ZIO E, BARALDI P. Evolutionary fuzzy clustering for the classification of transients in nuclear components[J]. Progress in Nuclear Energy, 2005, 46(3-4): 282-296. doi: 10.1016/j.pnucene.2005.03.010 [7] 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 [8] 艾鑫,刘永阔,蒋利平,等. 基于iForest-Adaboost的核电厂一回路故障诊断技术研究[J]. 核动力工程,2020, 41(3): 208-213. [9] 陈玉昇,杨燕华,林萌,等. 基于主元分析法的核反应堆关键参数提取研究[J]. 核动力工程,2019, 40(S2): 35-38. [10] LIU Y K, PENG M J, XIE C L, et al. Research and design of distributed fault diagnosis system in nuclear power plant[J]. Progress in Nuclear Energy, 2013, 68: 97-110. doi: 10.1016/j.pnucene.2013.06.002 [11] PENG B S, XIA H, LIU Y K, et al. Research on intelligent fault diagnosis method for nuclear power plant based on correlation analysis and deep belief network[J]. Progress in Nuclear Energy, 2018, 108: 419-427. doi: 10.1016/j.pnucene.2018.06.003 [12] LEE D, SEONG P H, KIM J. Autonomous operation algorithm for safety systems of nuclear power plants by using long-short term memory and function-based hierarchical framework[J]. Annals of Nuclear Energy, 2018, 119: 287-299. doi: 10.1016/j.anucene.2018.05.020 [13] ZHOU D X. Theory of deep convolutional neural networks: downsampling[J]. Neural Networks, 2020, 124: 319-327. doi: 10.1016/j.neunet.2020.01.018 -
计量
- 文章访问数: 308
- HTML全文浏览量: 101
- PDF下载量: 63
- 被引次数: 0
