FesNet for solving PDE which based on finite-element in Seismic

Model

Train

Dataset Para:
    super_element_dim = 294,
    super_step_size = 1000,
    super_step_t = 0.5, 
    super_hot_start = 1/35, 
    super_size=31,
    path = "./data/"
    

Train para:
    Adam,StepLR
    lr = 0.002
    hidden_dim = 400
    logits_size = 10
    num_layers = 2
    epoch = 10
    max_length = 10
    clip_threh = 0.5
    step_size = 5
    gamma = 0.5

Predict

Test_dataset: Time Cost: 5.15300178527832 s Time Cost: 3.7200005054473877 s Time Cost: 3.778003692626953 s Time Cost: 4.040002107620239 s Time Cost: 3.366999626159668 s Time Cost: 3.7520222663879395 s Time Cost: 3.792003870010376 s Time Cost: 3.3139984607696533 s

    Teacher-Forcing   Self-Regression
MSE : 4.136940627009608, 4.143258782220073
R2  : 0.6857                  0.6852

demo: Perfect one:

Bad one:

piglaker/fesnet

FesNet for solving PDE which based on finite-element in Seismic

Model

Train

Predict