Real-time Prediction of Dynamic Trend for Main Coolant Pump in Nuclear Power Plants
-
摘要: 准确的预测是核动力装置状态监测和运行维护的基础，为了提高系统和部件动态趋势预测的精度，本文提出了一种基于信号分解策略的时间序列预测方法。首先，利用变分模态分解（VMD）将原始时间序列信号分解为2个分别含有高频成分和低频趋势项的子序列。然后，采用贝叶斯优化（BOA）的门控循环单元（BOA-GRU）模型和自回归移动平均（ARIMA）模型分别对高频和低频子序列进行预测。最后，将2个子序列的预测值进行重构得到原始信号的预测结果。利用提出的混合模型对某核电厂主冷却剂泵的时间序列信号进行单步和多步预测，并利用均方根误差（RMSE）、平均绝对百分比误差（MAPE）、平均绝对误差（MAE）等指标对预测精度进行评估。结果表明，该混合模型能够对主冷却剂泵的运行状态进行准确地预测和追踪，并且与基础模型的对比突出了混合模型在复杂信号预测中的优势。
-
关键词:
- 变分模态分解（VMD） /
- 自回归移动平均（ARIMA） /
- 门控循环单元（GRU） /
- 混合模型；趋势预测 /
- 主冷却剂泵
Abstract: Accurate prediction is fundamental to the condition monitoring and operational maintenance of nuclear power plants (NPPs). To improve the dynamic trend prediction of systems and components, this paper proposes a time series prediction method based on signal decomposition strategy. Firstly, the original time series signal is decomposed into two subsequences containing high-frequency noise and low-frequency trend respectively, utilizing variational mode decomposition (VMD). Then, the gated recurrent unit optimized by the Bayesian optimization algorithm (BOA-GRU) and autoregressive integrated moving average (ARIMA) are employed to forecast these high-frequency and low-frequency subsequences separately. Finally, the predicted values of both subsequences are recombined to derive the forecast of the original signal. The proposed hybrid model is applied to perform single-step and multi-step predictions on the time series signals from the reactor coolant pump of a specific NPP, and the prediction accuracy is evaluated using metrics such as root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE). The results demonstrate that the proposed hybrid model can accurately predict and track the operational status of main coolant pump, and comparisons with baseline models highlight the advantages of the proposed hybrid model in complex signal prediction.-
Key words:
- VMD /
- ARIMA /
- GRU /
- Hybrid model,Trend prediction /
- Main coolant pump
-
表 1 GRU神经网络的超参数设置
Table 1. Hyperparameters of GRU Neural Networks
超参数 设置值或名称 GRU层数 1 梯度优化算法 自适应矩估计（Adam） L2正则化参数 0.001 训练迭代次数 500 嵌入维度 20 
下载: 导出CSV
表 2 不同模型的单步提前预测性能
Table 2. 1-step Ahead Prediction Performance of Models
信号 指标 混合模型 BOA-GRU模型 ARIMA模型 TS1 RMSE 0.0088 0.0112 0.0125 MAE 0.0070 0.0089 0.0098 MAPE/% 1.2692 1.5886 1.7435 TS2 RMSE 0.0159 0.0176 0.0168 MAE 0.0107 0.0122 0.0116 MAPE/% 2.2009 2.4415 2.3324
下载: 导出CSV
表 3 不同模型的5步提前预测性能
Table 3. 5-step Ahead Prediction Performance of Models
信号 指标 混合模型 BOA-GRU模型 ARIMA模型 TS1 RMSE 0.0123 0.0261 0.0161 MAE 0.0093 0.0194 0.0115 MAPE/% 1.6754 3.3754 2.0731 TS2 RMSE 0.0237 0.0376 0.0315 MAE 0.0171 0.0253 0.0181 MAPE/% 3.5686 5.0317 3.7119
下载: 导出CSV
-
[1] 朱少民,夏虹,彭彬森,等. 基于PCA的主泵传感器状态监测模型[J]. 核动力工程,2020, 41(3): 170-176. [2] SUN Y, HAGHIGHAT F, FUNG B C M. A review of the-state-of-the-art in data-driven approaches for building energy prediction[J]. Energy and Buildings, 2020, 221: 110022. doi: 10.1016/j.enbuild.2020.110022 [3] 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. [4] DEY R, SALEM F M. Gate-variants of gated recurrent unit (GRU) neural networks[C]//Proceedings of the 2017 IEEE 60th International Midwest Symposium on Circuits and Systems (MWSCAS). Boston: IEEE, 2017: 1597-1600. [5] 邱文智,张文煜,郭振海,等. 基于二次分解和乌鸦搜索算法优化组合模型的超短期风速预测[J]. 太阳能学报,2024, 45(3): 73-82. [6] DRAGOMIRETSKIY K, ZOSSO D. Variational mode decomposition[J]. IEEE Transactions on Signal Processing, 2014, 62(3): 531-544. doi: 10.1109/TSP.2013.2288675 [7] 崔佳旭,杨博. 贝叶斯优化方法和应用综述[J]. 软件学报,2018, 29(10): 3068-3090. [8] CHENG H Y, DING X W, ZHOU W N, et al. A hybrid electricity price forecasting model with Bayesian optimization for German energy exchange[J]. International Journal of Electrical Power & Energy Systems, 2019, 110: 653-666. [9] 张秀春,夏虹,刘永康,等. 基于移动窗核主成分分析的核电厂主泵故障检测[J]. 核动力工程,2024, 45(3): 234-240. -
下载:
图(7) / 表(3)
计量
- 文章访问数: 16
- HTML全文浏览量: 11
- PDF下载量: 0
- 被引次数: 0
