abelsr1710/navier-stokes-2d-fno
收藏资源简介:
该数据集是二维Navier-Stokes方程(粘度ν=1e-3)的模拟数据,基于Fourier Neural Operator(FNO)基准,用于粘性不可压缩流体在单位环面上的涡度形式演化。数据集包含1200个样本,每个样本表示涡度场w(x, y, t)在周期性64×64网格上20个时间步的时空演化,数据形状为[1200, 64, 64, 20](样本索引、x网格、y网格、时间)。数据集旨在用于神经算子和偏微分方程(PDE)代理模型的训练和评估,提供Parquet和PyTorch tensor两种格式,并建议用户自行分割为训练/验证/测试子集。典型用途包括使用前10个时间步作为输入预测后10个时间步,以支持时空预测任务。数据生成遵循原始FNO工作,采用随机初始条件和固定强迫项。
Simulations of the 2D Navier-Stokes equation for a viscous, incompressible fluid in vorticity form on the unit torus, following the Fourier Neural Operator (FNO) benchmark of Li et al. (2021). Each sample is the time evolution of the vorticity field w(x, y, t) on a periodic 64×64 grid over 20 time steps. The dataset is intended for training and testing of neural operators and PDE surrogate models, with 1200 trajectories in a single pool; users are expected to split it into train/validation/test subsets. Typical usage involves predicting future steps from past steps, such as using the first 10 steps as input to predict the remaining 10 steps. Data is available in Parquet and original PyTorch tensor formats.




