wgz_decision/models/tools.py

import torch


def test_one_episode(env, act, device):
    """get evaluate information, record the unbalance of after taking action"""
    record_system_info = []  # same as observation
    record_init_info = []  # include month,day,time
    env.TRAIN = False
    state = env.reset()
    record_init_info.append([env.month, env.day, env.current_time])
    print(f'current testing month is {env.month}, day is {env.day}')
    for i in range(24):
        s_tensor = torch.as_tensor((state,), device=device)
        a_tensor = act(s_tensor)
        action = a_tensor.detach().cpu().numpy()[0]
        state, next_state, reward, done = env.step(action)
        record_system_info.append([state[1], state[2], env.HST.current_soc(), env.HST.get_power(),
                                   env.EC.current_power, action, reward])
        state = next_state
    # add information of last step EC, HST.current_soc, HST.power, grid
    record_system_info[-1][2:5] = [env.final_step_outputs[0], env.final_step_outputs[1], env.final_step_outputs[2]]
    record = {'init_info': record_init_info, 'system_info': record_system_info}
    return record


def get_episode_return(env, act, device):
    episode_reward = 0.0  # sum of rewards in an episode
    episode_unbalance = 0.0
    state = env.reset()
    for i in range(24):
        s_tensor = torch.as_tensor((state,), device=device)
        a_tensor = act(s_tensor)
        action = a_tensor.detach().cpu().numpy()[0]  # not need detach(), because with torch.no_grad() outside
        state, next_state, reward, done, = env.step(action)
        state = next_state
        episode_reward += reward
        episode_unbalance += env.unbalance
        if done:
            break
    return episode_reward, episode_unbalance