Reliability Evaluation Method of Bolts of Reactor Internals in High-neutron-fluence-rate Region Based on XGBoost
-
摘要: 堆内构件螺栓长期处于高温高压以及高辐射环境，连接围板与成形板的螺栓存在辐照促应力腐蚀开裂（IASCC）现象。为提前预测螺栓在应力腐蚀环境下的剩余寿命，减少核电厂的备件库存，本文采用 XGBoost预测堆内构件螺栓在高辐照环境下的剩余寿命。首先，对压水堆高中子注量率区域的全周期剩余寿命演化数据进行分析处理，获得相关性模型；然后，提出基于数据驱动的XGBoost预测螺栓剩余寿命，该方法具有较强的泛化性与较高的准确率，可以很好地评估高中子注量率区域螺栓的可靠性；最后，以35000个样本作为训练集、15000个样本作为测试集，与国际原子能机构（IAEA）经验公式计算值比较，结果表明，XGBoost 预测准确率高达99.93%，优于多元线性回归方法和AdaBoost（使用线性损失函数/使用平方损失函数/使用指数损失函数）方法。Abstract: The bolts of reactor internals are in high temperature, high pressure and high radiation environment for a long time, and the bolts connecting the shroud and the forming plate are subject to irradiation assisted stress corrosion cracking (IASCC). In order to predict the service life of bolts in stress corrosion environment in advance and reduce the spare parts inventory of nuclear power plants, this paper uses the XGBoost(eXtreme Gradient Boosting) model to predict the service life of bolts of reactor internals in high irradiation environment. First, the whole-cycle residual life evolution data in the high-neutron-fluence-rate region of PWR are analyzed and processed, and the correlation model is obtained. Then, XGBoost model based on data driven is proposed to predict the residual life of bolts. This method has strong generalization and high accuracy, and can well evaluate the reliability of bolts in high-neutron-fluence-rate region; Finally, 35000 samples are used as the training set and 15000 samples as the test set, which are compared with the calculated values of the International Atomic Energy Agency (IAEA) empirical formula. The results show that the prediction accuracy of the XGBoost model is as high as 99.93%, which is better than the multiple linear regression method, AdaBoost (using a linear loss function), AdaBoost (using a squared loss function) and AdaBoost (using an exponential loss function) methods.
-
Key words:
- XGBoost /
- Life prediction /
- Reactor internals /
- Data driven
-
图 4 节点总数与准确率关系折线图
Figure 4. Line Chart of Relationship between Total Number of Nodes and Accuracy Rate
图 5 最大深度与准确率关系折线图
Figure 5. Line Chart of Relationship between Max. Depth and Accuracy Rate
图 6 学习率与准确率关系折线图
Figure 6. Line Chart of Relationship between Learning Rate and Accuracy Rate
表 1 5000次Monte Carlo直接抽样的概率分布
Table 1. Probability Distribution of 5000 Times of Monte Carlo Direct Sampling
参数名 分布类型 均值 标准差 中子注量率/( cm−2·h−1) 正态分布 1.75×1016 1.75×1014 压力/(N·mm−2) 正态分布 70.5 0.705 屈服强度/(N·mm−2) 正态分布 205 2.05 温度/K 正态分布 603.15 6.0315 
下载: 导出CSV
表 2 节点总数与对应准确率
Table 2. Total Number of Nodes and Corresponding Accuracy Rate
节点总数 10 20 30 40 50 准确率 0.2626 0.315648 0.3305 0.5801 0.7211 节点总数 100 200 300 400 450 准确率 0.9500 0.9977 0.9997 0.9998 0.9998 节点总数 500 550 600 650 800 准确率 0.9998 0.9998 0.9998 0.9998 0.9998
下载: 导出CSV
表 3 最大深度与对应准确率
Table 3. Max. Depth and Corresponding Accuracy Rate
最大深度 2 4 6 8 10 12 准确率 0.9972 0.9991 0.9996 0.9997 0.9997 0.9998 最大深度 14 16 18 20 22 准确率 0.9998 0.9998 0.9998 0.9998 0.9998
下载: 导出CSV
表 4 学习率与对应准确率
Table 4. Learning Rate and Corresponding Accuracy Rate
学习率 0.01 0.02 0.03 0.04 0.05 准确率 0.9933 0.9998 0.9998 0.9998 0.9998 学习率 0.06 0.07 0.08 0.09 0.10 准确率 0.9998 0.9998 0.9998 0.9998 0.9997
下载: 导出CSV
表 5 XGBoost部分实际值、部分预测值、部分单条数据准确率和整体平均准确率
Table 5. Partial Actual Value, Partial Predicted Value, Partial Single Data Accuracy Rate and Overall Average Accuracy Rate of XGBoost
数据序号 实际值/h 预测值/h 单条数据准确率 5828 316402 316188.2 0.9993 44461 329493 329831.9 0.9990 22862 307950 307488.1 0.9985 14027 315234 315204.7 0.9999 46918 325307 325410.7 0.9996 36567 306354 306382.7 0.9999 …… …… …… …… 49509 308087 307971.0 0.9996 15229 307001 307363.5 0.9988 39413 304068 304297.1 0.9992 39241 306424 306139.3 0.9991 33567 305332 305093.5 0.9992 整体平均准确率 0.9993
下载: 导出CSV
表 6 XGBoost与其他模型的对比
Table 6. Comparison of XGBoost with Other Models
模型 RMSE 整体平均准确率 XGBoost 100.57 0.9993 AdaBoost（使用线性损失函数） 672.93 0.9761 AdaBoost（使用平方损失函数） 605.46 0.9760 AdaBoost（使用指数损失函数） 688.02 0.9760 多元线性回归 146.70 0.9370
下载: 导出CSV
-
[1] 杨武. 辐照促进应力腐蚀破裂研究的进展[J]. 材料保护,1994, 27(2): 1-4. [2] ISHIHARA T. Corrosion failures and related research subjects in light water reactor power plants[J]. Zairyo-to-Kankyo, 1990, 39(11): 621-629. [3] 赵和安. 我国核电发展及其与紧固件的关系[J]. 兵器装备工程学报,2018, 39(4): 131-137. doi: 10.11809/bqzbgcxb2018.04.028 [4] 段远刚,赵晓刚,喻济兵. 压水堆堆内构件寿命管理中的关键问题[J]. 核动力工程,2003, 24(5): 401-404. doi: 10.3969/j.issn.0258-0926.2003.05.001 [5] 谢航,葛亮,蔡家藩. 核电站堆内构件围板螺栓的超声检测[J]. 无损检测,2020, 42(10): 49-51. doi: 10.11973/wsjc202010010 [6] 段远刚,许斌,唐传宝. 围板连接螺栓的辐照促进应力腐蚀裂纹研究[J]. 核动力工程,2007, 28(2): 62-65. doi: 10.3969/j.issn.0258-0926.2007.02.014 [7] 邓平,孙晨,彭群家,等. 堆芯结构材料辐照促进应力腐蚀开裂研究现状[J]. 中国腐蚀与防护学报,2015, 35(6): 479-487. [8] 李守彬,孔晨光,范岩成. 核电厂围板螺栓裂纹的失效分析与超声检测[J]. 无损检测,2019, 41(1): 58-60. doi: 10.11973/wsjc201901014 [9] WAS G S. Role of irradiation in stress corrosion cracking[M]//SICKAFUS K E, KOTOMIN E A, UBERUAGA B P. Radiation Effects in Solids. Dordrecht: Springer, 2007: 421-447. [10] 刘增瑞,徐杰,辛正高. “华龙一号”核反应堆围板和成形板组件的安装[J]. 装备机械,2018, 2(3): 16-21. doi: 10.3969/j.issn.1662-0555.2018.03.004 [11] 张振声,孙立斌,王海涛. HTR石墨堆内构件应力评价概率论和确定论方法比较[J]. 核动力工程,2011, 32(S1): 57-60. [12] PHAM X T T, HO T H. Using boosting algorithms to predict bank failure: an untold story[J]. International Review of Economics & Finance, 2021, 76: 40-54. [13] ZHU M L. Construction of quantization strategy based on random forest and XGBoost[C]//Proceedings of 2020 Conference on Artificial Intelligence and Healthcare. Taiyuan: ACM, 2020: 5-9. [14] LI C, CHEN Z Y, LIU J B, et al. Power load forecasting based on the combined model of LSTM and XGBoost[C]//Proceedings of 2019 the International Conference on Pattern Recognition and Artificial Intelligence. Wenzhou: ACM, 2019: 46-51. [15] 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. [16] CHOI H S, KIM S, OH J E, et al. XGBoost-based instantaneous drowsiness detection framework using multitaper spectral information of electroencephalography[C]//Proceedings of 2018 ACM International Conference on Bioinformatics, Computational Biology, and Health Informatics. Washington: ACM, 2018: 111-121. [17] ZHANG D, CHEN H D, ZULFIQAR H, et al. Developing and applying machine learning-based methods in special function protein identification[J]. Computational and Mathematical Methods in Medicine, 2021, 2021: 6664362. [18] 陈梅. 秦山核电库存控制探讨[J]. 企业技术开发,2015, 34(24): 170,172. [19] International Atomic Energy Agency. Assessment and management of ageing of major nuclear power plant components important to safety: PWR vessel internals: 2007 update[R]. Vienna: International Atomic Energy Agency, 2007. [20] CHEN W, LEI X X, CHAKRABORTTY R, et al. Evaluation of different boosting ensemble machine learning models and novel deep learning and boosting framework for head-cut gully erosion susceptibility[J]. Journal of Environmental Management, 2021, 284: 112015. doi: 10.1016/j.jenvman.2021.112015 -
计量
- 文章访问数: 135
- HTML全文浏览量: 30
- PDF下载量: 21
- 被引次数: 0
