energy = []
EE_rate_total = np.zeros(50)
EE_rate_mean = []
counter = 0
mean_min = 10
min_index = 0
for i_episode in range(1000):
print('iteration is %d' % i_episode)
# 获取回合 i_episode 第一个 observation
observation = env.reset()
observation_naive = env_naive.reset(bs_list=env.bs_list)
ep_r = 0
while count_time <= 1000000000:
count_time += 1
action = RL.choose_action(observation)
action_naive = env_naive.choose_action()
observation_, energy_cost, energy_cost_max, temp = env.step(action, beta=-10, lambda_=0.5) # 获取下一个 state
# print(observation_)
observation_naive_, energy_cost_naive, energy_cost_max_naive = env_naive.step(action_naive, temp, beta=-10)
index_ob = np.zeros(number)
for i in range(number):
index_ob[i] = i * 4 + 1
r1 = []
for i in index_ob:
r1.append(observation_[int(i)])
r1 = pd.DataFrame(r1)
# print(r1)
r1 = (r1 - 500) / 500.0
r1[r1.iloc[:, 0] != -1.0] = 0
from VCM_environment import VCMEN
from RL_brain import DuelingDQN
MEMORY_SIZE = 1000
ACTION_SPACE = 8
if __name__ == "__main__":
env = VCMEN()
load_model_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)),
"models")
agent = DuelingDQN(n_actions=ACTION_SPACE,
n_features=144,
memory_size=MEMORY_SIZE,
environment_name=env.name,
dueling=True,
load_model_dir=load_model_dir)
state_t, reward_t, win = env.observe()
step = 0
while not win:
step += 1
print(state_t)
# choose
observation = state_t.flatten()
action_t = agent.choose_action(observation)
# act
env.execute_action(action_t)
state_t_1, reward_t, win = env.observe()
state_t = state_t_1
print(step)