-
Notifications
You must be signed in to change notification settings - Fork 0
/
main_sac_async.py
187 lines (155 loc) · 8.35 KB
/
main_sac_async.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
import os
from GymEnvironments.environment_discrete_action_async import RelicEnv
import pandas as pd
from agents.SAC_discrete import SACAgent
from utils import order_state_variables, min_max_scaling, calculate_tank_soc
import numpy as np
import json
directory = os.path.dirname(os.path.realpath(__file__))
if __name__ == '__main__':
result_directory_path = 'D:\\OneDrive - Politecnico di Torino\\PhD_Silvio\\14_Projects\\002_PVZenControl\\Thermal_Electrical_Storage_Control\\'
result_directory = 'test_10'
safe_exploration = -1
discount_factor = 0.99
alpha = 0.05
tau = 0.005
automatic_entropy_tuning = False
learning_rate_actor = 0.0005
learning_rate_critic = 0.0005
n_hidden_layers = 2
n_neurons = 64
batch_size = 256
replay_buffer_capacity = 24 * 30 * 15
prediction_observations = ['electricity_price']
prediction_horizon = 0
min_temperature_limit = 10 # Below this value no charging
min_charging_temperature = 12 # Charging begins above this threshold
max_temperature_limit = 18 # Above this threshold no discharging
num_episodes = 50
result_directory_final = result_directory_path + result_directory
if not os.path.exists(result_directory_final):
os.makedirs(result_directory_final)
hidden_size = n_hidden_layers * [n_neurons]
config = {
'res_directory': result_directory_final,
# Change this folder to the path where you want to save the output of each episode
'weather_file': 'ITA_TORINO-CASELLE_IGDG',
'simulation_days': 90,
'tank_min_temperature': min_temperature_limit,
'tank_max_temperature': max_temperature_limit,
'price_schedule_name': 'electricity_price_schedule.csv',
'pv_nominal_power': 2000,
'battery_size': 2400}
env = RelicEnv(config)
# Import predictions
cooling_load_predictions = pd.read_csv('supportFiles\\prediction-cooling_load_perfect.csv')
electricity_price_predictions = pd.read_csv('supportFiles\\prediction-electricity_price_perfect.csv')
pv_power_generation_predictions = pd.read_csv('supportFiles\\prediction-pv_power_generation_perfect.csv')
electricity_price_schedule = pd.read_csv('supportFiles\\electricity_price_schedule.csv', header=None)
state_tank = ['time_of_day', 'day_of_week', 'electricity_price', 'storage_soc', 'battery_soc', 'cooling_load',
'pv_power_generation', 'outdoor_air_temperature']
state_battery = ['time_of_day', 'day_of_week', 'electricity_price', 'action_tank']
n_action_tank = 3
n_action_battery = 2
input_dims_tank = len(state_tank)
input_dims_battery = len(state_battery)
# define period for RBC control and
min_price = float(electricity_price_schedule[0].min())
min_temperature_limit = 10 # Below this value no charging
min_charging_temperature = 12 # Charging begins above this threshold
max_temperature_limit = 18 # Above this threshold no discharging
# evaluate SOC
max_storage_soc = calculate_tank_soc(min_temperature_limit, min_temperature_limit,
max_temperature_limit) # The storage is full
min_charging_storage_soc = calculate_tank_soc(min_charging_temperature, min_temperature_limit,
max_temperature_limit)
min_storage_soc = calculate_tank_soc(max_temperature_limit, min_temperature_limit,
max_temperature_limit) # The storage is empty
rbc_controller = None
# Initialize agent
agent = SACAgent(state_dim=input_dims,
action_dim=n_actions, hidden_dim=hidden_size, discount=discount_factor, tau=tau,
lr_critic=learning_rate_critic, lr_actor=learning_rate_actor,
batch_size=batch_size, replay_buffer_capacity=replay_buffer_capacity, learning_start=30*24,
reward_scaling=10., seed=0, rbc_controller=rbc_controller, safe_exploration=safe_exploration,
automatic_entropy_tuning=automatic_entropy_tuning, alpha=alpha)
# Define the number of episodes
score_history = []
# Training Loop
for episode in range(1, num_episodes + 1):
episode_step = 0
observation = env.reset(name_save='episode')
# append prediction
electricity_price = electricity_price_schedule[0][env.kStep + 1]
storage_soc = observation[3]
observation = order_state_variables(env_names=env.state_names,
observation=observation,
cooling_load_predictions=cooling_load_predictions,
electricity_price_predictions=electricity_price_predictions,
pv_power_generation_predictions=pv_power_generation_predictions,
horizon=prediction_horizon,
step=episode_step)
# Scale observations
observation = min_max_scaling(observation, env.state_mins, env.state_maxs, np.array([0]),
np.array([1]))
score = 0
done = False
actions_probabilities = []
q_values_1 = []
q_values_2 = []
while not done:
action = agent.choose_action(simulation_step=env.kStep + (episode - 1)*10000,
electricity_price=electricity_price,
storage_soc=storage_soc,
observation=observation)
step = 1
reward = 0
cooling_load = 0 # the cooling load needs to be averaged across simulation steps
auxiliary_load = 0
pv_power = 0
while step <= env.ep_time_step:
new_observation, reward_step, done, info = env.step(action)
cooling_load += new_observation[1]
auxiliary_load += new_observation[9]
pv_power += new_observation[8]
reward += reward_step
step += 1
cooling_load = cooling_load / env.ep_time_step # calculate average value
pv_power = pv_power / env.ep_time_step
auxiliary_load = auxiliary_load / env.ep_time_step
episode_step += 1
if done:
break
# append predictions
electricity_price = electricity_price_schedule[0][env.kStep + 1]
storage_soc = new_observation[3]
new_observation = list(new_observation)
new_observation[1] = cooling_load
new_observation[8] = pv_power
new_observation[9] = auxiliary_load
new_observation = tuple(new_observation)
new_observation = order_state_variables(env_names=env.state_names,
observation=new_observation,
cooling_load_predictions=cooling_load_predictions,
electricity_price_predictions=electricity_price_predictions,
pv_power_generation_predictions=pv_power_generation_predictions,
horizon=prediction_horizon,
step=episode_step)
# Scale observations
new_observation = min_max_scaling(new_observation, env.state_mins, env.state_maxs, np.array([0]),
np.array([1]))
# print(new_observation)
agent.remember(observation, action, reward, new_observation, done)
act_prob = agent.get_actions_probabilities(observation=new_observation)
actions_probabilities.append(act_prob.tolist())
score += reward
if episode != num_episodes:
agent.learn()
observation = new_observation
actions_probabilities = np.stack(actions_probabilities)
actions_probabilities = pd.DataFrame(actions_probabilities)
actions_probabilities.to_csv(
result_directory_final + '\\episode_{}_action_probabilities.csv'.format(str(episode)),
index=False, decimal=',', sep=';')
score_history.append(score)
print(f'Episode: {episode}, Score: {score}')