Skip to content

[feat] add code of prose_fd model#138

Open
lkyu-ly wants to merge 1 commit into
PaddlePaddle:developfrom
lkyu-ly:feat/prose_fd
Open

[feat] add code of prose_fd model#138
lkyu-ly wants to merge 1 commit into
PaddlePaddle:developfrom
lkyu-ly:feat/prose_fd

Conversation

@lkyu-ly

@lkyu-ly lkyu-ly commented Apr 15, 2026

Copy link
Copy Markdown

This PR integrates the Paddle version of prose_fd into PaddleCFD following the repository layout conventions:

  • Add core model/runtime code under ppcfd/models/prose_fd
  • Add training, evaluation, configs, and utilities under examples/prose_fd

Torch vs Paddle Forward Alignment

Torch vs Paddle forward alignment (shape: (1, 1, 128, 128, 4))

  • Abs diff -> max: 5.78e-06, mean: 3.75e-07, median: 2.98e-07
  • Rel diff -> max: 1.10e-01, mean: 9.03e-06, median: 1.20e-06
  • Rel diff -> p95: 1.37e-05, p99: 6.91e-05
  • Rel diff > 0.01 ratio: 0.01%
  • Rel diff > 0.001 ratio: 0.07%


## Data

The dataset we used are collected from [PDEBench](https://github.com/pdebench/PDEBench), [PDEArena](https://github.com/pdearena/pdearena), and [CFDBench](https://github.com/luo-yining/CFDBench). More details about data preprocessing are included in `data_utils/README.md`.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

PDEBench 这个数据集的所有数据都需要用到吗

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这个数据集太大了,因此复现和对齐使用的数据集是直接由PDEBench 源码生成的SWE数据。
为了减少数据大小,做出如下参数修改:

浅水方程 (SWE) — gen_radial_dam_break.py

  配置文件pdebench/data_gen/configs/radial_dam_break.yaml

  sim:
    n_time_steps: 100   # ← 时间步数(影响数据维度,保持100)
    xdim: 128           # ← 空间分辨率 X(改为64)
    ydim: 128           # ← 空间分辨率 Y(改为64)

  # 样本数:硬编码在脚本中
  # 文件:pdebench/data_gen/src/gen_radial_dam_break.py
  # 修改:num_samples_final = 10000  → 改为 200

注意num_samples_final  .py 脚本中硬编码需直接修改源文件

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants