56 lines
1.8 KiB
Python
56 lines
1.8 KiB
Python
import sys
|
|
import os
|
|
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
|
|
|
import torch
|
|
import torch.nn as nn
|
|
import config
|
|
from inverse.config_inverse import (
|
|
HIDDEN_SIZE, N_HIDDEN_LAYERS,
|
|
ALPHA_INIT, K_INIT, H_CONV_INIT,
|
|
)
|
|
|
|
|
|
class InverseHeatPINN(nn.Module):
|
|
def __init__(self, identify=('alpha', 'k', 'h_conv'), init_vals=None):
|
|
super().__init__()
|
|
layers = [nn.Linear(2, HIDDEN_SIZE), nn.Tanh()]
|
|
for _ in range(N_HIDDEN_LAYERS - 1):
|
|
layers += [nn.Linear(HIDDEN_SIZE, HIDDEN_SIZE), nn.Tanh()]
|
|
layers.append(nn.Linear(HIDDEN_SIZE, 1))
|
|
self.net = nn.Sequential(*layers)
|
|
self.identify = list(identify)
|
|
|
|
if init_vals is None:
|
|
init_vals = {}
|
|
|
|
_inits = {'alpha': ALPHA_INIT, 'k': K_INIT, 'h_conv': H_CONV_INIT}
|
|
_true = {'alpha': config.ALPHA, 'k': config.K, 'h_conv': config.H_CONV}
|
|
|
|
for name in ('alpha', 'k', 'h_conv'):
|
|
val = init_vals.get(name, _inits[name]) if name in identify else _true[name]
|
|
log_val = torch.log(torch.tensor(float(val)))
|
|
if name in identify:
|
|
setattr(self, f'log_{name}', nn.Parameter(log_val))
|
|
else:
|
|
self.register_buffer(f'log_{name}', log_val)
|
|
|
|
@property
|
|
def alpha(self):
|
|
return torch.exp(self.log_alpha)
|
|
|
|
@property
|
|
def k(self):
|
|
return torch.exp(self.log_k)
|
|
|
|
@property
|
|
def h_conv(self):
|
|
return torch.exp(self.log_h_conv)
|
|
|
|
def forward(self, xt):
|
|
x = xt[:, :1] / config.L
|
|
t = xt[:, 1:] / config.T_END
|
|
# Output scaling identica al forward PINN, ma usa i parametri appresi
|
|
T_char = config.T_AMB + (config.Q_VAL * config.L / self.k)
|
|
return config.T_AMB + (config.Q_VAL * config.L / self.k) * self.net(torch.cat([x, t], dim=1))
|