求解单层织物热湿耦合模型正反问题的物理信息神经网络方法
A Physics-Informed Neural Network Method for Solving Forward and Inverse Problems of Thermal-Humidity Coupling Model in Single-Layer Fabrics
蔡启凡 1徐映红2
作者信息
- 1. 浙江理工大学计算机科学与技术学院,浙江 杭州 310018
- 2. 浙江理工大学计算机科学与技术学院,浙江 杭州 310018;浙江理工大学理学院,浙江 杭州 310018
- 折叠
摘要
针对织物热湿耦合模型难以解耦和反问题求解时间长的问题,提出了一种求解单层稳态织物热湿耦合传递模型正反问题的物理信息神经网络(Physics-Informed Neural Networks,PINNs)方法.首先,给出了求解单层织物热湿传递方程正问题的PINNs方法,并采用数值实验验证了方法的有效性.其次,提出了基于热湿舒适性的厚度参数决定反问题,并使用PINNs方法进行求解.数值实验结果显示,PINNs方法在求解参数决定反问题时,仅需5 min即可预测出概率函数,相比于微分方程数值求解和粒子群结合方法,求解效率提高了25倍,展现出显著的优越性和应用潜力.
Abstract
In response to the challenges of decoupling the thermal-humidity coupling model of fabrics and the long solution time of inverse problems,this paper proposes a Physics-Informed Neural Networks(PINNs)method for solving the forward and inverse problems of the steady-state thermal-humidity coupling transfer model of single-layer fabrics.First,the PINNs approach is presented for solving the forward problem of the thermal-humidity transfer equation for single-layer fabrics,and numerical experiments are conducted to validate the effectiveness of the method.Next,an inverse problem determined by thickness parameters is proposed based on thermal-humidity comfort,which is solved using the PINNs method.The numerical experiment results show that the PINNs method can predict the probability function in just 5 minutes when solving the parameter-determined inverse problem,achieving a 25-fold increase in computational efficiency,compared to traditional numerical solutions of differential equations combined with particle swarm optimization methods,demonstrating significant advantages and application potential.
关键词
单层织物/热湿模型/耦合方程/神经网络/PINNsKey words
single-layer fabrics/thermal-humidity model/coupling equations/neural network/PINNs引用本文复制引用
出版年
2025
NETL
NSTL
万方数据