Fault Diagnosis Method of Nuclear Power Plant Based on Adaboost Algorithm
-
摘要: 针对当前基于集成学习的核电站故障诊断算法大多注重提高各种机器学习算法识别精度而忽略底层基学习器整合方法,导致集成学习算法识别事故类型精度难以提高,而且存在识别结果是否可信的问题。本文基于Adaboost算法设计了一种可使核电站控制系统自主识别故障类型的机器学习算法模型,该算法模型通过为集成学习的各种故障识别算法合理分配权重系数,提升集成学习整体算法对核电站事故类型的识别精度和算法可靠性。同时测试结果表明Adaboost算法对7种典型的核电站运行或事故工况的平均识别正确率可达95%以上;而且当事故发生150 s后,识别正确率可达100%。因此基于Adaboost算法的基学习器整合方法可用于优化集成学习的算法结构,提高算法对核电站事故类型的识别精度。
-
关键词:
- 核电站瞬态运行分析 /
- 故障诊断 /
- 机器学习 /
- Adaboost算法
Abstract: At present, most of the nuclear power plant fault diagnosis algorithms based on ensemble learning pay attention to improving the identification accuracy of various machine learning algorithms, while ignoring the integration method of the underlying base learner, which makes it difficult to improve the accuracy of the ensemble learning algorithm in identifying accident types, and there is a problem of whether the identification results are credible. In this paper, based on Adaboost algorithm, a machine learning algorithm model is designed to enable the control system of a nuclear power plant to identify fault types independently. By reasonably allocating weight coefficients for various fault identification algorithms of ensemble learning, the algorithm model improves the identification accuracy and reliability of the whole ensemble learning algorithm for nuclear power plant accident types. At the same time, the test results show that the average identification accuracy of Adaboost algorithm for seven typical nuclear power plant operation or accident conditions can reach more than 95%; And when the accident occurs 150 seconds, the identification accuracy can reach 100%. Therefore, the integration method of Adaboost algorithm to the base learner can be used to optimize the algorithm structure of ensemble learning and improve the identification accuracy of the algorithm for the types of nuclear power plant accidents. -
表 1 Adaboost算法对各种运行工况识别的混淆矩阵和精度表
Table 1. The Confusion Matrix and Precision Table of Adaboost Algorithm for Identification of Various Operating Conditions
运行工况 真正例 假反例 假正例 真反例 识别正确率/% 工况1 822 178 0 6000 97.45 工况2 766 234 116 5884 95.00 工况3 913 87 77 5923 97.65 工况4 773 227 68 5932 95.79 工况5 737 263 1 5999 96.23 工况6 737 265 6 5994 96.12 工况7 922 78 44 5956 98.26 -
[1] HOLMBERG J E, KAHLBOM U. Application of human reliability analysis in the deterministic safety analysis for nuclear power plants[J]. Reliability Engineering & System Safety, 2020, 194: 106371. [2] EEM S, CHOI I K, CHA S L, et al. Seismic response correlation coefficient for the structures, systems and components of the Korean nuclear power plant for seismic probabilistic safety assessment[J]. Annals of Nuclear Energy, 2021, 150: 107759. doi: 10.1016/j.anucene.2020.107759 [3] CHOUDRI M Z. AP1000非能动安全壳冷却系统整合确定论与概率论安全分析[D]. 北京: 华北电力大学(北京), 2019. [4] 张力,王春波,邹衍华,等. 核电厂安全相关人员行为确定论筛选方法研究[J]. 安全与环境学报,2018, 18(5): 1854-1858. [5] 黄志超,初永越,李虎伟,等. 风险指引型PSA应用的不确定性分析方法研究[J]. 核科学与工程,2017, 37(5): 830-838. doi: 10.3969/j.issn.0258-0918.2017.05.019 [6] SONG G S, KIM M C. Mathematical formulation and analytic solutions for uncertainty analysis in probabilistic safety assessment of nuclear power plants[J]. Energies, 2021, 14(4): 929. doi: 10.3390/en14040929 [7] ČEPIN M. Application of shutdown probabilistic safety assessment[J]. Reliability Engineering & System Safety, 2018, 178: 147-155. [8] LEE H, JEONG H, KOO G, et al. Attention recurrent neural network-based severity estimation method for interturn short-circuit fault in permanent magnet synchronous machines[J]. IEEE Transactions on Industrial Electronics, 2021, 68(4): 3445-3453. doi: 10.1109/TIE.2020.2978690 [9] XING S B, LEI Y G, WANG S H, et al. Distribution-invariant deep belief network for intelligent fault diagnosis of machines under new working conditions[J]. IEEE Transactions on Industrial Electronics, 2021, 68(3): 2617-2625. doi: 10.1109/TIE.2020.2972461 [10] 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 [11] ZHAO Y F, TONG J J, ZHANG L G. Rapid source term prediction in nuclear power plant accidents based on dynamic bayesian networks and probabilistic risk assessment[J]. Annals of Nuclear Energy, 2021, 158: 108217. doi: 10.1016/j.anucene.2021.108217 [12] RÄTSCH G, ONODA T, MÜLLER K R. Soft margins for AdaBoost[J]. Machine Learning, 2001, 42(3): 287-320. doi: 10.1023/A:1007618119488 [13] MERLER S, CAPRILE B, FURLANELLO C. Parallelizing AdaBoost by weights dynamics[J]. Computational Statistics & Data Analysis, 2007, 51(5): 2487-2498. [14] 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. [15] SRA S, NOWOZIN S, WRIGHT S J. Optimization for machine learning[M]. Cambridge: MIT Press, 2012. [16] QI K L, LI Q, LIU W, et al. Qinshan 300MWe NPP full scope simulator upgrade[R]. Vienna: International Atomic Energy Agency, 2006. [17] FAN J, UPADHYE S, WORSTER A. Understanding receiver operating characteristic (ROC) curves[J]. Canadian Journal of Emergency Medicine, 2006, 8(1): 19-20. doi: 10.1017/S1481803500013336