Research on Interpretable Diagnosis Method of Reactor Accidents Based on Representation Extraction
-
摘要: 为实现准确且可信的反应堆鲁棒事故诊断，本文构建了一种基于表征抽取且具有可解释性的诊断框架：首先提出了降噪遮掩自动编码器（DPAE）深度学习模型，DPAE在不同破口位置和破口大小的模拟数据集上进行自监督学习后，其编码器结构能够从部分缺失数据和噪声数据中自动提取监测参数的低维表征向量，进而将该表征向量用于基于分类和回归算法的下游诊断任务中；随后提出了一种基于后验可解释性算法的参数重要性计算方法，以分析监测参数对诊断结果的贡献。本研究以HPR1000为研究对象，在冷却剂丧失事故（LOCA）工况下验证了所提出的诊断方法。实验结果显示，在信噪比为30 dB的高斯噪声以及遮掩比例为0.3的随机遮掩干扰下，经训练的DPAE模型依然能获得有效的数据表征。此外，在受到信噪比为20 dB、遮掩比例为0.2的干扰下，相较于“端到端”诊断模型，本研究提出的“上下游”诊断模型在破口位置和尺寸诊断方面表现更优，并能识别对诊断结果贡献较大的监测参数。本研究提出的反应堆事故诊断方法有助于打造精确、稳定且可靠的智能反应堆运行维护系统。
-
关键词:
- 事故诊断 /
- 表征抽取 /
- 后验可解释性 /
- 降噪遮掩自动编码器（DAPE）
Abstract: This study proposes a diagnostic framework that is interpretable and based on representation extraction to achieve accurate, robust and reliable reactor accident diagnosis. Firstly, the Denoising Padded AutoEncoder (DPAE) deep learning model is introduced. Through self-supervised learning on a simulated dataset with varying break sizes and positions, the DPAE encoder can automatically extract low-dimensional representation vectors of monitoring parameters from partially missing data and noise data, which can then be used for downstream diagnostic tasks that involve classification and regression algorithms. Then, a parameter importance calculation method based on posteriori interpretability algorithm is introduced to analyze the contribution of monitoring parameters to diagnostic results. The proposed diagnostic method is validated using HPR1000 as the research object under LOCA conditions. The experimental results show that the trained DPAE model can still obtain effective data representation under Gaussian noise with signal-to-noise ratio of 30 dB and random masking ratio of 0.3. In addition, under the interference of signal-to-noise ratio of 20 dB and masking ratio of 0.2, compared with the "end-to-end" diagnosis model, the "upstream and downstream" diagnosis model proposed in this study performs better in the diagnosis of break position and size, and can identify monitoring parameters with greater contributions to the diagnostic results. The reactor accident diagnosis method proposed in this study is helpful to build an accurate, stable and reliable intelligent reactor operation and maintenance system. -
图 2 “上下游”式和“端到端”式诊断流程的差异
P—条件概率；C—类别参数；c—具体类别； f—函数映射；enc—编码过程；cls—分类器；reg—回归器
Figure 2. Differences between Upstream/downstream and End-to-end Diagnostic Processes
图 4 加入干扰后的样本流形分布特征
Figure 4. Manifold Distribution of Samples after Adding Interference
图 5 从受干扰样本中抽取的表征向量的流形分布特征
Figure 5. Manifold Distribution of Representation Vectors Drawn from Interfered Samples
图 6 表征向量元素对诊断输出影响的SHAP值分布
Figure 6. Distribution of SHAP Values for the Effect of Representation Vectors on Diagnostic Results
图 7 实验工况范围内输入参数对诊断的重要度排序
Figure 7. Importance Ranking of Input Parameters for Diagnosis in the Range of Experimental Conditions
图 8 对诊断结果重要度排名前五的监测参数
Figure 8. Top Five Monitoring Parameters in Terms of Importance to Diagnostic Results
表 1 不同诊断方法和模型的表现
Table 1. Performance of Different Diagnostic Methods and Models
诊断类型 算法 诊断准确率 F1 RMSE 冷腿 LOCA 热腿 LOCA “上下游” 多层感知机 0.905 0.934 0.919 0.312 支持向量机 0.898 0.917 0.908 0.649 XGBoost 0.911 0.946 0.927 0.564 随机森林 0.858 0.871 0.864 0.982 “端到端” TRES-CNN 0.682 0.646 0.662 2.945 BPNN 0.612 0.588 0.598 3.623 CNN 0.641 0.675 0.656 3.259 
下载: 导出CSV
-
[1] LIN L Y, ATHE P, ROUXELIN P, et al. Development and assessment of a nearly autonomous management and control system for advanced reactors[J]. Annals of Nuclear Energy, 2021, 150: 107861. doi: 10.1016/j.anucene.2020.107861 [2] AYODEJI A, LIU Y K, XIA H. Knowledge base operator support system for nuclear power plant fault diagnosis[J]. Progress in Nuclear Energy, 2018, 105: 42-50. doi: 10.1016/j.pnucene.2017.12.013 [3] LEE G, LEE S J, LEE C. A convolutional neural network model for abnormality diagnosis in a nuclear power plant[J]. Applied Soft Computing, 2021, 99: 106874. doi: 10.1016/j.asoc.2020.106874 [4] WANG H, PENG M J, HINES J W, et al. A hybrid fault diagnosis methodology with support vector machine and improved particle swarm optimization for nuclear power plants[J]. ISA Transactions, 2019, 95: 358-371. doi: 10.1016/j.isatra.2019.05.016 [5] LI J K, LIN M. Ensemble learning with diversified base models for fault diagnosis in nuclear power plants[J]. Annals of Nuclear Energy, 2021, 158: 108265. doi: 10.1016/j.anucene.2021.108265 [6] MA J P, JIANG J. Applications of fault detection and diagnosis methods in nuclear power plants: A review[J]. Progress in Nuclear Energy, 2011, 53(3): 255-266. doi: 10.1016/j.pnucene.2010.12.001 [7] JIANG B T, ZHOU J, HUANG X B. Artificial neural networks in condition monitoring and fault diagnosis of nuclear power plants: a concise review[C]//International Conference on Nuclear Engineering. American Society of Mechanical Engineers, 2020, 83778: V002T08A032. [8] HU G, ZHOU T T, LIU Q F. Data-driven machine learning for fault detection and diagnosis in nuclear power plants: a review[J]. Frontiers in Energy Research, 2021, 9: 663296. doi: 10.3389/fenrg.2021.663296 [9] MAITLO A A, LIU Y K, LAKHAN M N, et al. Recent advances in nuclear power plant for fault detection and diagnosis-a review[J]. Journal of Critical Reviews, 2020, 7(18): 4340-4349. [10] CHEN T Q, GUESTRIN C. XGBoost: a scalable tree boosting system[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco: ACM, 2016: 785-794. [11] LUNDBERG S M, LEE S I. A unified approach to interpreting model predictions[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach: Curran Associates Inc., 2017: 4768-4777. [12] ZHOU J Y, WU D, DING S H, et al. Research on analysis and modeling methods for LBLOCA in nuclear power plants based on autonomous LOCA analysis platform ARSAC[C]// International Conference on Nuclear Engineering.Los Angeles: American Society of Mechanical Engineers, 2020, 83785: V003T12A047 [13] BREIMAN L. Random forests[J]. Machine Learning, 2001, 45(1): 5-32. doi: 10.1023/A:1010933404324 [14] CORTES C, VAPNIK V. Support-vector networks[J]. Machine Learning, 1995, 20(3): 273-297. [15] BASU A, BARTLETT E B. Detecting faults in a nuclear power plant by using dynamic node architecture artificial neural networks[J]. Nuclear Science and Engineering, 1994, 116(4): 313-325. doi: 10.13182/nse94-a18990 [16] Li C, Li M, Qiu Z. Post-hoc Interpretability Based Parameter Selection for Data Oriented Nuclear Reactor Accident Diagnosis System[C]//International Conference on Nuclear Engineering. Beijing: American Society of Mechanical Engineers, 2022, 86427: V07BT07A014. [17] 李航. 统计学习方法[M]. 第二版. 北京: 清华大学出版社, 2019: 28-30. [18] VAN DER MAATEN L, HINTON G. Visualizing data using t-SNE[J]. Journal of Machine Learning Research, 2008, 9(86): 2579-2605. [19] SHAPLEY L S. A value for n-person games[J]. Contributions to the Theory of Games,1953, 2(28):307-317. -
计量
- 文章访问数: 220
- HTML全文浏览量: 107
- PDF下载量: 83
- 被引次数: 0
