-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathdata_preprocessing.py
More file actions
198 lines (163 loc) · 9.58 KB
/
data_preprocessing.py
File metadata and controls
198 lines (163 loc) · 9.58 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
import h5py
import numpy as np
import torch
import matplotlib.pyplot as plt
plt.set_cmap('jet')
def prepare_data3d(Ksteps, data_dir, state_data, ctrl_data, yobs_data, cond):
#### Load pressure and saturation data
pho_co2 = 1.8696 ### kg/m^3
pho_water = 1000 ### kg/m^3
hf_r = h5py.File(data_dir + state_data, 'r')
mole = np.array(hf_r.get('Mole_frac_norm_slt')).transpose((3,2,1,0))
sat = np.array(hf_r.get('Sg_norm_slt')).transpose((3,2,1,0))
pres = np.array(hf_r.get('Psim_norm_slt')).transpose((3,2,1,0))
hf_r.close()
n_sample, steps_slt, Ny, Nx = mole.shape
print(mole.shape)
# plt.imshow(sat[0,0,:,:])
#### Load output data
hf_r = h5py.File(data_dir + yobs_data)
# Qmax_w = torch.tensor(np.array(hf_r.get('Q_max_w')), dtype=torch.float32)
# Qmax_g = torch.tensor(np.array(hf_r.get('Q_max_g')), dtype=torch.float32)
# Qmax_RC = torch.tensor(np.array(hf_r.get('Q_max_RC')), dtype=torch.float32)
# if cond =='RC':
# Qrate_w = np.array(hf_r.get('Qpro_w_RC_norm_slt')).transpose((2,1,0))
# Qrate_g = np.array(hf_r.get('Qpro_g_RC_norm_slt')).transpose((2,1,0))
# else:
# Qrate_w = np.array(hf_r.get('Qpro_w_norm_slt')).transpose((2,1,0))
# # Qrate_w = Qrate_w*Qmax_w*pho_water*360/1000/1000000
# Qrate_g = np.array(hf_r.get('Qpro_g_norm_slt')).transpose((2,1,0))
# # Qrate_g = Qrate_g*Qmax_g*pho_co2*360/1000/1000000 #### convert original normalized data to mass rate in MMT/yr
BHP_inj = np.array(hf_r.get('BHPinj_norm_slt')).transpose((2,1,0))
hf_r.close()
# yobs = np.concatenate((Qrate_w,Qrate_g,BHP_inj),axis=1)
# yobs = np.concatenate((Qrate_w,BHP_inj),axis=1)
yobs = BHP_inj
#### Load control data
hf_r = h5py.File(data_dir + ctrl_data)
# bhp0 = np.array(hf_r.get('Pwf_norm_slt')).transpose((2,1,0))
rate0 = np.array(hf_r.get('Qinj_norm_slt')).transpose((2,1,0))
# rate0 = rate0*Qrate_g*pho_co2*360/1000/1000000 #### convert original normalized data to mass rate in MMT/yr
hf_r.close()
# bhp = np.concatenate((bhp0,rate0),axis=1)
bhp = rate0
print(bhp.shape)
n_sample, num_well, steps_ctrl = bhp.shape
# _, num_prod, _ = bhp0.shape
_, num_inj, _ = rate0.shape
Mole_slt = []
SAT_slt = []
PRES_slt = []
BHP_slt = []
Yobs_slt = []
indt = np.array(range(0,steps_slt-(Ksteps-1)))
print(indt)
for k in range(Ksteps):
indt_k = indt + k
if k ==1:
indt_del = indt_k - indt
indt_del = indt_del / max(indt_del)
mole_t_slt = mole[:, indt_k,:, :]
sat_t_slt = sat[:, indt_k,:, :]
pres_t_slt = pres[:, indt_k,:, :]
num_t_slt = sat_t_slt.shape[1]
if k < Ksteps-1:
bhp_t_slt = np.swapaxes(bhp[:,:, indt_k],1,2)
yobs_t_slt = np.swapaxes(yobs[:,:, indt_k],1,2)
Mole_slt.append(mole_t_slt)
SAT_slt.append(sat_t_slt)
PRES_slt.append(pres_t_slt)
if k < Ksteps-1:
BHP_slt.append(bhp_t_slt)
Yobs_slt.append(yobs_t_slt)
return Mole_slt, SAT_slt, PRES_slt, BHP_slt, Yobs_slt, num_t_slt, Nx, Ny, num_well, 0, num_inj
def train_split_data(Mole_slt, SAT_slt, PRES_slt, BHP_slt, Yobs_slt, num_t_slt, Nx, Ny, num_well, num_prod, num_inj, n_channels, device):
num_all = Mole_slt[0].shape[0]
split_ratio = int(num_all/100)
# num_run_per_case = 75
# num_run_eval = 100 - num_run_per_case # 25 cases
num_run_per_case = 75
num_run_eval = 100 - num_run_per_case # 20 cases
mole_t_train = np.zeros((num_run_per_case*split_ratio, num_t_slt, Ny, Nx))
sat_t_train = np.zeros((num_run_per_case*split_ratio, num_t_slt, Ny, Nx))
pres_t_train = np.zeros((num_run_per_case*split_ratio, num_t_slt, Ny, Nx))
bhp_t_train = np.zeros((num_run_per_case*split_ratio, num_t_slt, num_well))
# yobs_t_train = np.zeros((num_run_per_case*split_ratio, num_t_slt, 2*num_prod+num_inj))
yobs_t_train = np.zeros((num_run_per_case*split_ratio, num_t_slt, num_prod+num_inj))
mole_t_eval = np.zeros((num_run_eval*split_ratio, num_t_slt, Ny, Nx))
sat_t_eval = np.zeros((num_run_eval*split_ratio, num_t_slt, Ny, Nx))
pres_t_eval = np.zeros((num_run_eval*split_ratio, num_t_slt, Ny, Nx))
bhp_t_eval = np.zeros((num_run_eval*split_ratio, num_t_slt, num_well))
# yobs_t_eval = np.zeros((num_run_eval*split_ratio, num_t_slt, 2*num_prod+num_inj))
yobs_t_eval = np.zeros((num_run_eval*split_ratio, num_t_slt, num_prod+num_inj))
num_train = num_run_per_case*split_ratio*num_t_slt
shuffle_ind_train = np.random.default_rng(seed=1010).permutation(num_train)
num_eval = num_run_eval*split_ratio*num_t_slt
shuffle_ind_eval = np.random.default_rng(seed=1010).permutation(num_eval)
STATE_train = []
BHP_train = []
Yobs_train = []
STATE_eval = []
BHP_eval = []
Yobs_eval = []
for i_step in range(len(SAT_slt)):
for k in range(split_ratio):
ind0 = k * num_run_per_case
mole_t_train[ind0:ind0+num_run_per_case,...] = Mole_slt[i_step][k*100:k*100+num_run_per_case,...]
sat_t_train[ind0:ind0+num_run_per_case,...] = SAT_slt[i_step][k*100:k*100+num_run_per_case,...]
pres_t_train[ind0:ind0+num_run_per_case,...] = PRES_slt[i_step][k*100:k*100+num_run_per_case,...]
if i_step<len(SAT_slt)-1:
bhp_t_train[ind0:ind0+num_run_per_case,...] = BHP_slt[i_step][k*100: k*100+num_run_per_case,...]
yobs_t_train[ind0:ind0+num_run_per_case,...] = Yobs_slt[i_step][k*100: k*100+num_run_per_case,...]
# dt_train[ind0:ind0+num_run_per_case,...] =indt_d_slt[k*100: k*100+num_run_per_case, :, :]
# Eval set
ind1 = k*num_run_eval
mole_t_eval[ind1:ind1+num_run_eval,...] = Mole_slt[i_step][k*100+num_run_per_case:k*100+100,...]
sat_t_eval[ind1:ind1+num_run_eval,...] = SAT_slt[i_step][k*100+num_run_per_case:k*100+100,...]
pres_t_eval[ind1:ind1+num_run_eval,...] = PRES_slt[i_step][k*100+num_run_per_case:k*100+100,...]
if i_step<len(SAT_slt)-1:
bhp_t_eval[ind1:ind1+num_run_eval,...] = BHP_slt[i_step][k*100+num_run_per_case: k*100+100,...]
yobs_t_eval[ind1:ind1+num_run_eval,...] = Yobs_slt[i_step][k*100+num_run_per_case: k*100+100,...]
Mole_t_train = mole_t_train.reshape((num_run_per_case*split_ratio*num_t_slt, 1, Ny, Nx))
SAT_t_train = sat_t_train.reshape((num_run_per_case*split_ratio*num_t_slt, 1, Ny, Nx))
PRES_t_train = pres_t_train.reshape((num_run_per_case*split_ratio*num_t_slt, 1, Ny, Nx))
if i_step<len(SAT_slt)-1:
BHP_t_train = bhp_t_train.reshape((num_run_per_case*split_ratio*num_t_slt, num_well))
# Yobs_t_train = yobs_t_train.reshape((num_run_per_case*split_ratio*num_t_slt, 2*num_prod+num_inj))
Yobs_t_train = yobs_t_train.reshape((num_run_per_case*split_ratio*num_t_slt, num_prod+num_inj))
# DT_train = dt_train.reshape((num_run_per_case*4*num_t_slt, 1))
Mole_t_eval = mole_t_eval.reshape((num_run_eval*split_ratio*num_t_slt, 1, Ny, Nx))
SAT_t_eval = sat_t_eval.reshape((num_run_eval*split_ratio*num_t_slt, 1, Ny, Nx))
PRES_t_eval = pres_t_eval.reshape((num_run_eval*split_ratio*num_t_slt, 1, Ny, Nx))
if i_step<len(SAT_slt)-1:
BHP_t_eval = bhp_t_eval.reshape((num_run_eval*split_ratio*num_t_slt, num_well))
# Yobs_t_eval = yobs_t_eval.reshape((num_run_eval*split_ratio*num_t_slt, 2*num_prod+num_inj))
Yobs_t_eval = yobs_t_eval.reshape((num_run_eval*split_ratio*num_t_slt, num_prod+num_inj))
# DT_eval = dt_eval.reshape((num_run_eval*4*num_t_slt, 1))
### shuffle train and eval samples
if n_channels==3:
STATE_t_train = torch.tensor(np.concatenate((Mole_t_train, SAT_t_train, PRES_t_train),axis=1), dtype=torch.float32).to(device)
STATE_t_eval = torch.tensor(np.concatenate((Mole_t_eval, SAT_t_eval, PRES_t_eval),axis=1), dtype=torch.float32).to(device)
else:
STATE_t_train = torch.tensor(np.concatenate((Mole_t_train, PRES_t_train),axis=1), dtype=torch.float32).to(device)
STATE_t_eval = torch.tensor(np.concatenate((Mole_t_eval, PRES_t_eval),axis=1), dtype=torch.float32).to(device)
STATE_t_train = STATE_t_train[shuffle_ind_train, ...]
if i_step<len(SAT_slt)-1:
BHP_t_train = torch.tensor(BHP_t_train[shuffle_ind_train, ...], dtype=torch.float32).to(device)
Yobs_t_train = torch.tensor(Yobs_t_train[shuffle_ind_train, ...], dtype=torch.float32).to(device)
# DT_train = DT_train[shuffle_ind_train, ...]
STATE_t_eval = STATE_t_eval[shuffle_ind_eval, ...]
if i_step<len(SAT_slt)-1:
BHP_t_eval = torch.tensor(BHP_t_eval[shuffle_ind_eval, ...], dtype=torch.float32).to(device)
Yobs_t_eval = torch.tensor(Yobs_t_eval[shuffle_ind_eval, ...], dtype=torch.float32).to(device)
# DT_eval = DT_eval[shuffle_ind_eval, ...]
STATE_train.append(STATE_t_train)
STATE_eval.append(STATE_t_eval)
if i_step<len(SAT_slt)-1:
BHP_train.append(BHP_t_train)
BHP_eval.append(BHP_t_eval)
Yobs_train.append(Yobs_t_train)
Yobs_eval.append(Yobs_t_eval)
return STATE_train, BHP_train, Yobs_train, STATE_eval, BHP_eval, Yobs_eval
def save_data_to_file():
return