/
main.py
290 lines (238 loc) · 10.9 KB
/
main.py
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
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
import os
import argparse
from collections import OrderedDict
from pathlib import Path
import torch
import torch.nn as nn
from torch.utils.tensorboard import SummaryWriter
from citylearn import CityLearn
from agent import RL_Agents
from utils.io import get_output_folder
from reward_function import reward_function
from utils.standardization import normalize_seq2seq_state_forRL, normalize_state
MAX_EPISODES = 300
sim_period = 8760
parser = argparse.ArgumentParser()
# RL Hyper-parameters
parser.add_argument('--lr', type=float, default=1e-4)
parser.add_argument('--MAX_BUFFER', type=int, default=10000)
parser.add_argument('--seed', '-s', type=int, default=0)
parser.add_argument('--BATCH_SIZE', type=int, default=32)
parser.add_argument('--climate_zone', type=int, default=2)
parser.add_argument('--act_limit', type=float, default=0.5)
parser.add_argument('--decay', type=float, default=1)
# TCN Hyper-parameters
parser.add_argument('--encode_dim', type=int, default=64)
parser.add_argument('--alpha', type=float, default=0.2)
parser.add_argument('--kernel_size', type=int, default=3)
parser.add_argument('--kernel_size_forecast', type=int, default=3)
parser.add_argument('--levels', type=int, default=5)
parser.add_argument('--levels_forecast', type=int, default=3)
parser.add_argument('--seq_len', type=int, default=24)
parser.add_argument('--seq_len_forecast', type=int, default=6)
# reward Hyper-parameters
parser.add_argument('--A_r', type=float, default=0)
parser.add_argument('--B_r', type=float, default=1)
parser.add_argument('--window_len_A', type=int, default=6)
parser.add_argument('--window_len_B', type=int, default=12)
parser.add_argument('--price_factor', type=float, default=0.01)
# logger
parser.add_argument('--print_per_step', type=int, default=1000)
parser.add_argument('--writing_per_step', type=int, default=1000)
parser.add_argument('--start_k', type=int, default=0)
parser.add_argument('--start_c', type=int, default=0)
# load model
parser.add_argument('--continue_flag', type=int, default=0)
parser.add_argument('--load_episode', type=int, default=295)
args = parser.parse_args()
reward_kwargs = OrderedDict(
alpha=args.A_r,
beta=args.B_r,
total_energy_window=args.window_len_A,
heat_energy_window=args.window_len_B,
price_factor=args.price_factor,
ramping_window=6
)
full_dim = 33
src_dim = 21
pred_dim = 4
latent_dim = 3
def split_state_mixedTCN(norm_states):
# (batch, building, seq, full_dim + src_dim + pred_dim)
return norm_states[:, :, :, 0:full_dim], \
norm_states[:, :, :, full_dim:full_dim + src_dim], \
norm_states[:, :, -args.seq_len_forecast:, -pred_dim:]
def split_state_TCN(norm_states, future_len=6):
src = norm_states[:, :, :, 0:src_dim]
if future_len == 6:
# (batch, building, seq, src_dim + 4)
return src, norm_states[:, :, -args.seq_len_forecast:, -pred_dim:]
elif future_len == 12:
# (batch, building, seq, src_dim + 4(pred12) + 4(pred6))
pred12hr = norm_states[:, :, -6:, -2*pred_dim:-pred_dim]
pred6hr = norm_states[:, :, -6:, -pred_dim:]
pred = torch.cat((pred6hr, pred12hr), -2)
return src, pred
# Load Model using args dict
def get_agent_kwargs(args, log_path, eval=True):
# Lazy Import
from model.BaseModules import ActionClipping
from model.Encoder import ROMAEncoder
from model.RLModules import DualHyperQNet, SGHNActor, MLP
encode_dim = args.encode_dim
encoder_kwargs = (ROMAEncoder,
OrderedDict( # for ROMAEncoder
input_size=33,
output_size=64,
role_size=latent_dim)
)
model_kwargs = OrderedDict(
Encoder=encoder_kwargs,
DualCritic=(DualHyperQNet,
OrderedDict(input_size=encode_dim + 2, latent_size=latent_dim)),
Actor=(SGHNActor,
OrderedDict(input_size=encode_dim, output_size=2, latent_size=latent_dim)),
DisparityNet=(MLP,
OrderedDict(input_size=3, output_size=1, layer_sizes=[64],
norm_layer=nn.BatchNorm1d, activation=nn.LeakyReLU())),
)
default_lr = args.lr
action_clip_kwargs = OrderedDict(
start_bound=args.act_limit,
stop_bound=.1,
decay=args.decay,
step_size=20000,
warm_up=0,
verbose=True
)
# print("act limit:", action_clip_kwargs['start_bound)
if eval:
algo_kwargs = None
print("eval mode, use deterministic policy")
else:
unique_optim_kwargs = OrderedDict(
Encoder=OrderedDict(),
DualCritic=OrderedDict(),
Actor=OrderedDict()
)
algo_kwargs = OrderedDict(
batch_size=args.BATCH_SIZE,
alpha=args.alpha,
buffer_capacity=args.MAX_BUFFER,
optim_cls=torch.optim.Adam,
optim_kwargs=OrderedDict(lr=default_lr, betas=(0.9, 0.999), eps=1e-8, weight_decay=0),
unique_optim_kwargs=unique_optim_kwargs,
verbose=True,
log_interval=args.print_per_step,
log_path=log_path
)
ac_kwargs = dict(
model_kwargs=model_kwargs,
algo_kwargs=algo_kwargs,
# state_fn=lambda s: normalize_seq2seq_state_forRL(s, future_len=args.seq_len_forecast),
state_fn=normalize_state,
# No reward_fn for Hierarchical Agents # TODO Compatibility with original agent
reward_fn=lambda rewards, states: reward_function(rewards, states, **reward_kwargs),
action_clipping=lambda model: ActionClipping(model, **action_clip_kwargs),
memory_size=args.seq_len
)
return ac_kwargs
def init_pretrain_models(rl_agents, seq2seq_path='Models_seq2seq_withAE_future'+str(args.seq_len_forecast),
autoencoder_path='Models_one_AE_128dim'):
rl_agents.encoder.history_AE.load_state_dict(torch.load('pretrained_models/AE/{}_zone{}/AE.pt'.format(autoencoder_path, args.climate_zone)))
rl_agents.encoder.history_AE.eval()
rl_agents.encoder.history_AE.requires_grad_(False)
rl_agents.encoder.auto_encoder.load_state_dict(torch.load('pretrained_models/AE_noSOC/{}_noSOC_zone{}/AE.pt'.format(autoencoder_path, args.climate_zone)))
rl_agents.encoder.auto_encoder.eval()
rl_agents.encoder.auto_encoder.requires_grad_(False)
rl_agents.encoder.seq2seq.load_state_dict(torch.load('pretrained_models/seq2seq/{}_zone{}/Seq2SeqLSTM.pt'.format(seq2seq_path, args.climate_zone)))
rl_agents.encoder.seq2seq.eval()
rl_agents.encoder.seq2seq.requires_grad_(False)
if __name__ == "__main__":
filename = "zone_" + str(args.climate_zone) + \
"_lr" + str(args.lr) + \
"_predLen_" + str(args.seq_len_forecast) + \
"_windowlenB_" + str(args.window_len_B) + \
"_actlimit" + str(args.act_limit) + \
"_A_" + str(args.A_r) + \
"_B_" + str(args.B_r) + \
"_decay_" + str(args.decay)
# Instantiating the Tensorboard writers
PATH_base = 'datas/new/'
PATH_base = get_output_folder(PATH_base, 'scalar_' + filename)
PATH_to_log_dir1 = PATH_base + '/reward'
writer = SummaryWriter(PATH_to_log_dir1)
PATH_to_log_dir2 = PATH_base + '/cost'
cost_writer = SummaryWriter(PATH_to_log_dir2)
ac_kwargs = get_agent_kwargs(args, PATH_base, eval=False)
# load data
data_path = Path("../data/Climate_Zone_" + str(args.climate_zone))
building_attributes = data_path / 'building_attributes.json'
weather_file = data_path / 'weather_data.csv'
solar_profile = data_path / 'solar_generation_1kW.csv'
building_state_actions = 'buildings_state_action_space.json'
building_ids = ["Building_1", "Building_2", "Building_3", "Building_4", "Building_5", "Building_6", "Building_7",
"Building_8", "Building_9"]
objective_function = ['ramping', '1-load_factor', 'average_daily_peak', 'peak_demand',
'net_electricity_consumption', 'total']
# Instantiating the env
env = CityLearn(data_path, building_attributes, weather_file, solar_profile, building_ids,
buildings_states_actions=building_state_actions, cost_function=objective_function)
observations_spaces, actions_spaces = env.get_state_action_spaces()
# Provides information on Building type, Climate Zone, Annual DHW demand, Annual Cooling Demand,
# Annual Electricity Demand, Solar Capacity, and correllations among buildings
building_info = env.get_building_information()
# Select many episodes for training. In the final run we will set this value to 1 (the buildings run for one year)
start_episode = 0
# RL CONTROLLER
# Instantiating the control agent(s)
rl_agent = RL_Agents(building_info, observations_spaces, actions_spaces, ac_kwargs)
k, c = args.start_k, args.start_c
cost = {}
# initialize the model
best_model_path = "./Models_best_zone" + str(args.climate_zone)
# pretrained part
# init_pretrain_models(rl_agent.agent)
# print("load pretrain models successfully.")
# non-pretrained part
if args.continue_flag:
rl_agent.agent.load_models(best_model_path, args.load_episode)
print("load model from ", args.load_episode)
model_path = './Models_' + filename
if not os.path.isdir(model_path):
os.mkdir(model_path)
# print(rl_agent.agent.encoder)
# The number of episodes can be replaces by a stopping criterion (i.e. convergence of the average reward)
for e in range(0, MAX_EPISODES):
cum_reward = {}
for id in range(rl_agent.n_buildings):
cum_reward[id] = 0
state = env.reset()
done = False
while not done:
if k % (40000 * 4) == 0:
print('hour: ' + str(k) + ' of ' + str(sim_period * MAX_EPISODES))
action = rl_agent.select_action(state)
next_state, reward, done, _ = env.step(action)
rl_agent.add_to_buffer(state, action, reward, next_state, done)
state = next_state
for id in range(rl_agent.n_buildings):
cum_reward[id] += reward[id]
if k % args.writing_per_step == 0:
print("write rewards per " + str(args.writing_per_step) + "step")
for id in range(rl_agent.n_buildings):
writer.add_scalar('building_reward_' + str(id), reward[id], k)
k += 1
# write episode-accumulated reward
for r in range(rl_agent.n_buildings):
writer.add_scalar('building_cum_reward_' + str(r), cum_reward[r], c)
# write cost
cost[e] = env.cost()
print("cost_writer adding scalar")
for i in range(len(objective_function)):
cost_writer.add_scalar(str("cost_") + objective_function[i], cost[e][objective_function[i]], c)
cost_writer.flush()
# save models
if c % 1 == 0:
rl_agent.agent.save_models(model_path, e)
c += 1