基于PINN的燃料棒稳态温度分布快速预测方法研究  

Research on Fast Prediction Method of Fuel Rod Steady-state Temperature Distribution Based on PINN

作　　者：刘振海[1] 张涛 齐飞鹏 张坤[1] 李垣明[1] 周毅[1] 李文杰[1] Liu Zhenhai;Zhang Tao;Qi Feipeng;Zhang Kun;Li Yuanming;Zhou Yi;Li Wenjie(Science and Technology on Reactor System Design Technology Laboratory,Nuclear Power Institute of China,Chengdu,610213,China)

机构地区：[1]中国核动力研究设计院核反应堆系统设计技术重点实验室,成都610213

出　　处：《核动力工程》2024年第S01期39-44,共6页Nuclear Power Engineering

摘　　要：本研究建立了一种基于物理信息神经网络(PINN)的燃料棒稳态温度分布快速预测方法。将燃耗、线功率、温度边界、空间位置等作为特征参数,利用PINN求解参数化的固体导热方程。基于该方法分别建立了燃料芯块和包壳稳态温度分布快速预测模型,计算结果表明:快速预测模型的计算速度相比商业有限元软件而言快1000倍,同时具有较高精度,芯块和包壳稳态温度与验证集相比预测最大相对偏差分别约0.318%、0.013%,可以快速且准确地预测燃料棒稳态温度分布。A fast prediction method of fuel rod steady-state temperature distribution base on Physical Informed Neural Network(PINN)is established in this research.The burnup,linear power,boundary temperature and space position are taken as characteristic parameters to solve the parametric solid heat conduction equations using PINN.Based on this method,rapid prediction models for the steady-state temperature distribution of fuel pellet and cladding were constructed.The calculation results show that the calculation speed of fast prediction models are about 1000 times faster than that of commercial finite element method software,and they also have high accuracy.The maximum relative deviation of the steady-state temperature prediction of fuel pellets and cladding is about 0.318%and 0.013%respectively compared with the validation set.The established PINN model can quickly and accurately predict the steady-state temperature distribution of fuel rods.

关 键 词：物理信息神经网络(PINN) 燃料棒稳态温度 快速预测 燃料行为 

分 类 号：TL334[核科学技术—核技术及应用]