相位保持频带路由傅里叶算子

林 冠泽, 史 晨熙, 王靖 涵*
大连外国语大学

摘要


针对固定频谱权重难以表征样本间尺度响应差异、幅相拆分又易损伤结构信息的问题，提出相位保持自适
应频带路由傅里叶神经算子（PABR-FNO）。该方法以低、中、高频归一化能量驱动轻量路由器，在固定频谱预算
内生成样本级实值门控，并同步缩放复系数实部与虚部；同时引入点式空间残差以补偿局部非线性。64点周期多尺
度合成算子的8个配对种子实验表明，PABR-FNO的相对L2误差为0.1666±0.0150，较基础FNO降低52.70%；精确
Wilcoxon检验经Holm校正后p=0.03125。参数匹配消融显示，自适应路由和复相位保留均为必要组成。结论仅适用
于结构匹配的无噪声机制仿真。

关键词


傅里叶神经算子；频带路由；相位保持；多尺度学习；算子学习

全文:

PDF


参考


[1]LI Z, KOVACHKI N, AZIZZADENESHELI K, et

al. Fourier neural operator for parametric partial differential

equations[C/OL]//International Conference on Learning

Representations. [S.l.]: OpenReview.net, 2021[2026-08-12].

[2]GUIBAS J, MARDANI M, LI Z, et al. Adaptive

Fourier neural operators: efficient token mixers for

transformers[C/OL]//International Conference on Learning

Representations. [S.l.]: OpenReview.net, 2022[2026-08-12].

[3]TRAN A, MATHEWS A, XIE L, et al. Factorized

Fourier neural operators[C/OL]//International Conference

on Learning Representations. [S.l.]: OpenReview.net,

2023[2026-08-12].

[4]WEN G, LI Z, AZIZZADENESHELI K, et al.

U-FNO— An enhanced Fourier neural operator-based

deep-learning model for multiphase flow[J]. Advances in

Water Resources, 2022, 163:104180.

[5]LI X Y, YANG Y B. GAFNO: gated adaptive

Fourier neural operator for task-agnostic time series

modeling[C]//2023 IEEE International Conference on Data

Mining. 2023:1133-1138.

[6]LI K, YE W. D-FNO: a decomposed Fourier

neural operator for large-scale parametric partial differential

equations[J]. Computer Methods in Applied Mechanics and

Engineering, 2025, 436:117732.


Refbacks

  • 当前没有refback。