Fourier 神经算子:给仿真做“代理模型”
Fourier Neural Operator for Parametric Partial Differential Equations
Li et al. · 2020 · arXiv:2010.08895
Foundation paper · permanent
2020
能否学一个“算子”,一次训练后对整族参数直接出解,而不必每次重跑求解器?
What they did在频域做全局卷积,捕捉长程相关;分辨率无关。
Evidence在 Navier–Stokes 等参数化 PDE 上,误差显著低于此前的神经算子。
Do not over-trust外推到训练参数区间之外会迅速失效;用它替代校核,等于把验证环节删掉。
Limits需要成百上千组高保真样本;不保证守恒律;边界复杂时精度下降。
In-class task生成 300 组工况训练代理,在区间内外各取 20 组,比较误差放大倍数。
A foundation paper is the landmark work for a method and stays permanently: it explains why the method works, and is never replaced for being old.
© SailiTech · Printed at: