Research on Prediction Method of Reactor Axial Power Deviation Based on Combined Feature Selection and Temporal Convolutional Network
-
摘要: 反应堆轴向功率偏差能够反映堆芯轴向功率分布和反应堆的运行情况，针对轴向功率偏差在变工况下预测困难的问题，该文提出一种基于组合特征筛选与时序卷积网络（TCN）的反应堆轴向功率偏差预测方法。以轴向功率偏差控制的基本原则为出发点，分析影响轴向功率偏差变化的因素，综合分析多维特征间的冗余度与相关性，利用组合特征筛选策略形成面向轴向功率偏差预测的最优特征子集，构建面向轴向功率偏差预测的关键关联特征数据，输入至TCN捕捉动态因果关系，以实现反应堆轴向功率偏差预测。实验研究表明，该文所提轴向功率偏差预测方法可深度挖掘反应堆轴向功率偏差相关参量的时序因果变化特性，准确预测轴向功率偏差发展态势，解决传统预测模型在复杂工况下预测跟踪不及时的问题，对核电厂反应堆状态监测和安全运行提供辅助参考的依据。
-
关键词:
- 参量预测 /
- 轴向功率偏差 /
- 时序卷积网络（TCN） /
- 组合特征筛选
Abstract: The axial power deviation of a reactor can reflect the axial power distribution of the core and the operation of the reactor. Aiming at the difficulties in predicting the axial power deviation under variable operating conditions, this paper proposes a prediction method of reactor axial power deviation based on the combined feature selection and temporal convolutional network (TCN). Taking the basic principle of axial power deviation control as the starting point, this paper analyzes the factors affecting the change of axial power deviation, comprehensively analyzes the redundancy and correlation among multi-dimensional features, uses the combined feature selection strategy to form the optimal feature subset for axial power deviation prediction, constructs the key correlation feature data for axial power deviation prediction, and inputs it into TCN to capture dynamic causality, so as to achieve the prediction of reactor axial power deviation. Experimental studies show that the proposed method can deeply explore the temporal causal change characteristics of the parameters related to the axial power deviation of the reactor, accurately predict the development trend of the axial power deviation, solve the problem that the traditional prediction model does not predict and track in time under complex operating conditions, and provide an auxiliary reference basis for the reactor status monitoring and safe operation of nuclear power plants. -
图 1 基于组合特征筛选与TCN的轴向功率偏差预测框架
Figure 1. Framework for Axial Power Deviation Prediction Based on Combined Feature Selection and TCN
图 3 因果卷积结构（左）和扩张卷积结构（右）
Figure 3. Structure of Causal Convolution (left) and Structure of Dilated Causal Convolution (right)
图 8 卷积核大小、学习率参数寻优
Figure 8. Parameter Optimization of Convolution Kernel Size and Learning Rate
图 9 有/无最优特征子集15步∆I预测结果
Figure 9. 15-step ∆I Prediction Results with/without Optimal Feature Subset
图 10 不同模型∆I预测结果与误差对比
Figure 10. Comparison of Prediction Results and Error by Different Models for ∆I
表 1 初始关联特征集描述
Table 1. Description of Initial Association Feature Set
特征符号 特征含义 特征符号 特征含义 特征符号 特征含义 ∆I 轴向功率偏差 R7 控制棒7棒位 f12 控制棒2棒位变化率 I1 第1段探测器电流 R8 控制棒8棒位 f13 控制棒3棒位变化率 I2 第2段探测器电流 f1 轴向功率偏差变化率 f14 控制棒4棒位变化率 I3 第3段探测器电流 f2 第1段探测器电流变化率 f15 控制棒5棒位变化率 I4 第4段探测器电流 f3 第2段探测器电流变化率 f16 控制棒6棒位变化率 I5 第5段探测器电流 f4 第3段探测器电流变化率 f17 控制棒7棒位变化率 I6 第6段探测器电流 f5 第4段探测器电流变化率 f18 控制棒8棒位变化率 R1 控制棒1棒位 f6 第5段探测器电流变化率 Ia1 上半部探测器电流平均值 R2 控制棒2棒位 f7 第6段探测器电流变化率 Ia2 下半部探测器电流平均值 R3 控制棒3棒位 f8 上半部探测器电流变化率 Ia3 各段探测器电流平均值 R4 控制棒4棒位 f9 下半部探测器电流变化率 Id1 上、下半部探测器电流之差 R5 控制棒5棒位 f10 上、下半部探测器电流差变化率 Rs1 所有控制棒棒位之和 R6 控制棒6棒位 f11 控制棒1棒位变化率 Rs2 所有控制棒棒位变化率之和 
下载: 导出CSV
表 2 有效特征预测结果
Table 2. Prediction Results of Effective Features
特征 δRMSE/%FP 第1次 第2次 第3次 平均值 误差提升量 ∆I 0.1800 0.2023 0.1633 0.1818 Rs1 0.0763 0.1137 0.0523 0.0808 0.1011 R2 0.0925 0.0968 0.1003 0.0965 0.0853 R6 0.1170 0.1443 0.0859 0.1157 0.0661 R7 0.0952 0.1204 0.1368 0.1175 0.0644 R3 0.1510 0.1221 0.1013 0.1248 0.0570 R4 0.1206 0.1178 0.1553 0.1312 0.0506 R5 0.1514 0.1437 0.0993 0.1315 0.0504 R1 0.1614 0.1463 0.1133 0.1403 0.0415 R8 0.1858 0.1542 0.0988 0.1463 0.0356 f9 0.1649 0.1410 0.1904 0.1654 0.0164 f4 0.1868 0.1507 0.1778 0.1718 0.0101 f5 0.1805 0.1768 0.1625 0.1733 0.0086 f10 0.1740 0.1739 0.1793 0.1757 0.0061 f8 0.1880 0.1532 0.1971 0.1794 0.0024 f3 0.1661 0.1860 0.1883 0.1801 0.0017 f6 0.1841 0.1570 0.2003 0.1805 0.0013
下载: 导出CSV
表 3 TCN重要超参数设定
Table 3. Important Hyperparameter Settings of TCN
参数描述 设定值 批处理大小 128 学习率 0.005 卷积核大小 5 训练迭代次数 50 各卷积层卷积核个数 {16,32,16} 优化器 Adam
下载: 导出CSV
表 4 有/无最优特征子集效果对比
Table 4. Effect Comparison with/without Optimal Feature Subset
预测
结果误差类型 δRMSE/%FP δMAPE/% 预测步数 5 10 15 5 10 15 整体 有S* 0.048 0.070 0.088 1.105 1.652 1.878 无S* 0.069 0.087 0.127 1.473 1.672 2.037 局部 有S* 0.058 0.081 0.106 1.362 1.869 2.479 无S* 0.107 0.178 0.299 2.209 3.530 5.724
下载: 导出CSV
-
[1] 舒印彪,赵勇,赵良,等. “双碳”目标下我国能源电力低碳转型路径[J]. 中国电机工程学报,2023, 43(5): 1663-1671. [2] 董德英,张琳. 核保护系统响应时间自动测试系统设计[J]. 核电子学与探测技术,2020, 40(3): 504-507. doi: 10.3969/j.issn.0258-0934.2020.03.025 [3] 常晓露. 压水堆核电站轴向功率控制原理与方法[J]. 科技视界,2021(7): 88-89. [4] 陈自强,陈成. 压水堆核电厂寿期末堆芯轴向功率偏差控制[J]. 核科学与工程,2012, 32(S1): 63-70. [5] 胡晓英. 轴向偏差报警系统设计优化及KMP仪表调试问题研究[J]. 仪器仪表用户,2014, 21(1): 19-21. doi: 10.3969/j.issn.1671-1041.2014.01.004 [6] 陈静,卢燕臻,江灏,等. 基于孪生模型的堆芯自给能中子探测器信号异常检测[J]. 核动力工程,2023, 44(3): 210-216. [7] 林蔚青,缪希仁,陈静,等. 时空信息融合的堆芯自给能中子探测器故障检测与隔离方法[J]. 中国电机工程学报,2024, 44(23): 9310-9322. [8] 朱少民,夏虹,吕新知,等. 基于ARIMA和LSTM组合模型的核电厂主泵状态预测[J]. 核动力工程,2022, 43(2): 246-253. [9] 张思原,卢忝余,曾辉,等. 基于LSTM的核电传感器多特征融合多步状态预测[J]. 核动力工程,2021, 42(4): 208-213. [10] RACHEAL S, LIU Y K, AYODEJI A. Evaluation of optimized machine learning models for nuclear reactor accident prediction[J]. Progress in Nuclear Energy, 2022, 149: 104263. doi: 10.1016/j.pnucene.2022.104263 [11] KIM H, KIM J. Long-term prediction of safety parameters with uncertainty estimation in emergency situations at nuclear power plants[J]. Nuclear Engineering and Technology, 2023, 55(5): 1630-1643. doi: 10.1016/j.net.2023.01.026 [12] CHEN F K, DONG X M, LUO Y C, et al. Research on the prediction of LOCA condition in nuclear power plants based on GRU recurrent neural network and its variants[J]. Quality and Reliability Engineering International, 2024, 40(2): 759-782. doi: 10.1002/qre.3446 [13] FU Y J, ZHANG D Z, XIAO Y L, et al. An interpretable time series data prediction framework for severe accidents in nuclear power plants[J]. Entropy, 2023, 25(8): 1160. doi: 10.3390/e25081160 [14] SHI T Z, SHE J K, LI P E, et al. A deep learning-based framework for the operation prediction of primary heat transfer loop in nuclear power plants[J]. Frontiers in Energy Research, 2023, 11: 1099326. doi: 10.3389/fenrg.2023.1099326 [15] GONG H L, YU Y R, PENG X J, et al. A data-driven strategy for xenon dynamical forecasting using dynamic mode decomposition[J]. Annals of Nuclear Energy, 2020, 149: 107826. doi: 10.1016/j.anucene.2020.107826 [16] PARK J H, AN Y J, YOO K H, et al. Leak flow prediction during loss of coolant accidents using deep fuzzy neural networks[J]. Nuclear Engineering and Technology, 2021, 53(8): 2547-2555. doi: 10.1016/j.net.2021.01.040 [17] EJIGU D A, LIU X J. Gradient descent-particle swarm optimization based deep neural network predictive control of pressurized water reactor power[J]. Progress in Nuclear Energy, 2022, 145: 104108. doi: 10.1016/j.pnucene.2021.104108 [18] PEDERSEN K T. Predicting xenon oscillations in PWRs using intrusive reduced order modelling[D]. Gothenburg: Chalmers University of Technology, 2023. [19] ABDULRAHEEM K K, TOLOKONSKY A O, LAIDANI Z. Adaptive second-order sliding-mode control for a pressurized water nuclear reactor in load following operation with Xenon oscillation suppression[J]. Nuclear Engineering and Design, 2022, 391: 111742. doi: 10.1016/j.nucengdes.2022.111742 [20] 刘国发,郭文琪. 核电厂仪表与控制[M]. 北京: 中国原子能出版社,2010: 143-146. [21] RESHEF D N, RESHEF Y A, FINUCANE H K, et al. Detecting novel associations in large data sets[J]. Science, 2011, 334(6062): 1518-1524. doi: 10.1126/science.1205438 [22] ZHANG Z C, HONG W C. Application of variational mode decomposition and chaotic grey wolf optimizer with support vector regression for forecasting electric loads[J]. Knowledge-Based Systems, 2021, 228: 107297. doi: 10.1016/j.knosys.2021.107297 -
计量
- 文章访问数: 38
- HTML全文浏览量: 11
- PDF下载量: 0
- 被引次数: 0
