Study on Transient Parameter Prediction and Fault Diagnosis of Nuclear Power Plant Based on LSTM Neural Network
-
摘要: 为提高核电厂瞬态工况下参数预测和故障诊断的准确性和实时性，本研究采用长短期记忆（LSTM）神经网络模型进行预测和诊断。通过生成并随机化故障情景，减少预测模型对特定模式的依赖，提高其在未知故障情景下的泛化能力。研究结合沙普利加性解释性（SHAP）方法，对预测模型的参数预测结果进行解释性分析，评估不同输入特征对模型预测性能的影响，并验证该预测模型在传感器故障和数据传输错误情况下的预测准确性。此外，针对含有不同噪声水平的瞬态参数进行故障诊断，验证故障诊断模型的鲁棒性。结果表明，LSTM神经网络模型在预测和诊断方面具有较高的精度，即使在传感器故障、数据传输有误以及数据含有噪声情况下仍表现出色。本研究提出的方法能够提升核电厂运行安全和稳定性，为事故工况下的安全性提供有效技术支持。
-
关键词:
- 核电厂 /
- 瞬态参数预测 /
- 故障诊断 /
- LSTM神经网络 /
- 沙普利加性解释性（SHAP）方法
Abstract: To improve the accuracy and real-time performance of parameter prediction and fault diagnosis under transient conditions in nuclear power plants, this study employs a Long Short-Term Memory (LSTM) neural network model for prediction and diagnosis. By generating and randomizing fault scenarios, the model’s dependence on specific patterns is reduced, and its generalization capability in unknown fault situations is enhanced. The study integrates SHAP (SHapley Additive exPlanations) to conduct interpretability analysis on the parameter prediction results, evaluates the impact of different input features on the model’s predictive performance, and verifies its effectiveness under sensor failures and data transmission errors. Furthermore, fault diagnosis is performed on transient parameters with different noise levels to validate the model's robustness. The results demonstrate that the LSTM model achieves high accuracy in both prediction and diagnosis, and it maintains excellent performance even under sensor failures, data transmission errors, and noisy data. The method proposed in this study can improve the safety and stability of nuclear power plant operation and provide effective technical support for safety under accident conditions. -
图 2 LSTM神经网络参数预测原理图
Figure 2. Schematic Diagram of LSTM Neural Network Parameter Prediction
图 3 LSTM神经网络预测模型与SVM预测模型下LOCA关键参数趋势预测对比图
Figure 3. Comparison of Trend Prediction of Key Parameters of LOCA under LSTM Prediction Model and SVM Prediction Model
图 4 LSTM神经网络预测模型与SVM预测模型下SGTR关键参数趋势预测对比图
Figure 4. Comparison of Trend Prediction of SGTR Key Parameters under LSTM Prediction Model and SVM Prediction Model
图 5 LSTM神经网络预测模型与SVM预测模型下RW关键参数趋势预测对比图
Figure 5. Comparison of Trend Prediction of RW Key Parameters under LSTM Prediction Model and SVM Prediction Model
图 6 LSTM神经网络预测模型与SVM预测模型下SLBIC关键参数趋势预测对比图
Figure 6. Comparison of Trend Prediction of SLBIC Key Parameters under LSTM Prediction Model and SVM Prediction Model
图 7 稳压器水位为实际1.1倍时LSTM神经网络预测模型关键参数运行趋势预测图
Figure 7. Predicted Trend of Key Parameters When Pressurizer Water Level is 1.1 Times the Actual Level
图 8 冷却剂平均温度缺失时LSTM神经网络预测模型关键参数运行趋势预测图
Figure 8. Predicted Trend of Key Parameters When Coolant Average Temperature Missing
图 11 使用预测参数进行故障诊断混淆矩阵图
Figure 11. Confusion Matrix for Fault Diagnosis Using Predicted Parameters
图 12 使用实际参数进行故障诊断混淆矩阵图
Figure 12. Confusion Matrix for Fault Diagnosis Using Actual Parameters
图 13 不同噪声水平下故障诊断混淆矩阵图
Figure 13. Confusion Matrix for Fault Diagnosis at Different Noise Levels
表 1 故障设置
Table 1. Fault Settings
编号 故障类型 故障情景设置 0 LOCA 破口面积：1~16 cm2 1 SGTR 传热管破裂率：1%~16% 2 RW 反应性增加范围：1% ~16% 3 SLBIC 破口面积：1~16 cm2 
下载: 导出CSV
表 2 LSTM神经网络预测模型超参数设置
Table 2. Model Hyperparameter Settings
参数描述 超参数设置 隐藏层神经元数目 64、54、256 激活函数 Sigmoid 优化器 Adam 迭代次数 10000 批处理大小 90 正则化技术 L2、Dropout 学习率 0.0004
下载: 导出CSV
表 3 LSTM神经预测模型与SVM预测模型预测效果
Table 3. Prediction Performance of Network Models
故障类型 模型类型 MAE RMSE LOCA LSTM 0.0053 0.0148 SVM 0.0206 0.0359 SGTR LSTM 0.0061 0.0198 SVM 0.0246 0.0258 RW LSTM 0.0051 0.0149 SVM 0.0240 0.0242 SLBIC LSTM 0.0076 0.0154 SVM 0.0355 0.0360
下载: 导出CSV
表 4 稳压器水位为实际数据1.1倍时LSTM神经网络预测模型性能
Table 4. Model Prediction Performance When Pressurizer Water Level is 1.1 Times the Actual Level
故障类型 MAE RMSE LOCA 0.0104 0.0206 SGTR 0.0062 0.0059 RW 0.0071 0.0265 SLBIC 0.0099 0.0230
下载: 导出CSV
表 5 冷却剂平均温度缺失时LSTM神经网络预测模型性能
Table 5. Model Prediction Performance When Coolant Average Temperature Missing
故障类型 MAE RMSE LOCA 0.0114 0.0124 SGTR 0.0084 0.0207 RW 0.0115 0.0327 SLBIC 0.0093 0.0197
下载: 导出CSV
表 6 LSTM神经网络故障诊断模型的性能
Table 6. Performance of the Fault Diagnosis Model
数据 准确率/% 精确率/% 召回率/% 预测参数 96.75 96.79 96.75 实际参数 99.50 99.50 99.50
下载: 导出CSV
表 7 添加噪声后LSTM神经网络故障诊断模型的性能
Table 7. Performance of Fault Diagnosis Model after Adding Noise
噪声水平/% 准确率/% 精确率/% 召回率/% 1 95.00 95.04 93.64 3 94.00 94.11 94.00 5 92.25 92.36 92.25
下载: 导出CSV
-
[1] 刘峰,余刃,李凤宇,等. 基于复合人工神经网络的核电站实时故障诊断方法研究[J]. 核动力工程,2007, 28(2): 95-100. doi: 10.3969/j.issn.0258-0926.2007.02.022 [2] WANG P F, ZHANG J X, WAN J S, et al. A fault diagnosis method for small pressurized water reactors based on long short-term memory networks[J]. Energy, 2022, 239: 122298. doi: 10.1016/j.energy.2021.122298 [3] BAE J, KIM G, LEE S J. Real-time prediction of nuclear power plant parameter trends following operator actions[J]. Expert Systems with Applications, 2021, 186: 115848. doi: 10.1016/j.eswa.2021.115848 [4] 冯千懿,郭张鹏,李仲春,等. 基于时序深度学习模型的安全壳关键参数快速预测研究[J]. 核动力工程,2022, 43(6): 79-84. [5] 孙原理,宋志浩. 基于卷积长短期记忆网络和人工鲸鱼算法的核反应堆运行事件诊断方法研究[J]. 核动力工程,2022, 43(4): 185-190. [6] 梁杰恒. 基于深度学习的核电厂运行状态智能诊断及预测方法研究[D]. 广州: 华南理工大学,2022. [7] HOCHREITER S. The vanishing gradient problem during learning recurrent neural nets and problem solutions[J]. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, 1998, 6(2): 107-116. doi: 10.1142/S0218488598000094 [8] KHAN A H, ISLAM M S. A pctran-based investigation on the effect of inadvertent control rod withdrawal on the thermal-hydraulic parameters of a VVER-1200 nuclear power reactor[J]. Acta Mechanica Malaysia, 2019, 2(2): 32-38. doi: 10.26480/amm.02.2019.32.38 [9] MAHMOUD T A, SHETA A A, FIKRY R M, et al. Design of data-driven model for the pressurizer system in nuclear power plants using a TSK fuzzy neural network[J]. Nuclear Engineering and Design, 2022, 399: 112015. doi: 10.1016/j.nucengdes.2022.112015 [10] 刘飞. 核电厂计算机化操作规程系统研究[D]. 哈尔滨: 哈尔滨工程大学,2009. [11] LUNDBERG S M, ERION G, CHEN H, et al. From local explanations to global understanding with explainable AI for trees[J]. Nature Machine Intelligence, 2020, 2(1): 56-67. doi: 10.1038/s42256-019-0138-9 [12] VASEGHI S V. Advanced digital signal processing and noise reduction[M]. Hoboken: John Wiley & Sons, 2008: 48-134. [13] WATANABE K, MATSUURA I, ABE M, et al. Incipient fault diagnosis of chemical processes via artificial neural networks[J]. AIChE Journal, 1989, 35(11): 1803-1812. doi: 10.1002/aic.690351106 [14] CAI L F, TIAN X M, CHEN S. A process monitoring method based on noisy independent component analysis[J]. Neurocomputing, 2014, 127: 231-246. doi: 10.1016/j.neucom.2013.07.029 -
计量
- 文章访问数: 49
- HTML全文浏览量: 17
- PDF下载量: 6
- 被引次数: 0
