摘要
为满足核应急对核素扩散分布快速预测的需求,本研究提出了一种基于TCN-BiLSTM混合神经网络模型的核素扩散浓度快速预测方法。通过时序卷积网络TCN提取时序特征,并利用双向长短期记忆网络BiLSTM捕捉双向时序依赖。采用真实的地形数据与气象数据,经拉格朗日扩散模型计算核素浓度分布数据,构成训练时序数据集。选取美国伊利诺伊州电厂SF6扩散示踪实验数据集Kincaid,验证TCN-BiLSTM模型的有效性。针对模拟137Cs泄漏事故为案例,模型预测值与拉格朗日模型计算结果偏差小于2%,与TCN相比MAE和RMSE分别降低了29.7%和33.3%,该方法可为核素扩散分布预测提供快速有效支持。
To meet the demand for rapid prediction of radionuclide dispersion in nuclear emergency response,this study proposes a hybrid TCN-BiLSTM model for fast concentration forecasting.Temporal features are extracted by the TCN,while bidirectional dependencies are captured by the BiLSTM.Training data are generated using realistic terrain and meteorological inputs with a Lagrangian dispersion model.The model is validated on the Kincaid SF6 tracer experiment dataset in Illinois,USA,and further applied to a simulated 137Cs release case.Results show that prediction errors remain below 2%,and compared with a standalone TCN,the hybrid model reduces MAE and RMSE by 29.7%and 33.3%,respectively,demonstrating its efficiency and accuracy for rapid radionuclide dispersion prediction.
作者
王箫箫
杨子辉
艾雨星
李煜辰
莫紫雯
袁振豪
李桃生
WANG Xiaoxiao;YANG Zihui;AI Yuxing;LI Yuchen;MO Ziwen;YUAN Zhenhao;LI Taosheng(College of Physics and Electronic Engineering,Huaibei Normal University,Huaibei 235000,China;Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230031,China;Science Island Bra nch,Uni versity of Science and Technology o f China,Hefei 230026,Ch ina)
出处
《辐射研究与辐射工艺学报》
2025年第6期102-113,共12页
Journal of Radiation Research and Radiation Processing
基金
安徽省自然科学基金(2008085MA23)
合肥物质科学研究院院长基金(YZJJ202208-TS)
科工局乏燃料后处理专项。
关键词
时序卷积网络
双向长短期记忆网络
核素大气扩散
浓度预测
Temporal convolutional network
Bidirectional long short-term memory
Radionuclide atmospheric dispersion
Concentration prediction
