Main Configuration File¶
The main environment for wrapping the environments is sustaindc_env.py
, which reads configurations from dc_config.json
and manages the external variables using managers for weather, carbon intensity, and workload.
Example Configuration¶
# Default configuration for this environment. New parameters should be
# added here
DEFAULT_CONFIG = {
# Agents active
'agents': ['agent_ls', 'agent_dc', 'agent_bat'],
# Datafiles
'location': 'ny',
'cintensity_file': 'NYIS_NG_&_avgCI.csv',
'weather_file': 'USA_NY_New.York-Kennedy.epw',
'workload_file': 'Alibaba_CPU_Data_Hourly_1.csv',
# Capacity (MW) of the datacenter
'datacenter_capacity_mw': 1,
# Maximum battery capacity
'max_bat_cap_Mw': 2,
# weight of the individual reward (1=full individual, 0=full collaborative, default=0.8)
'individual_reward_weight': 0.8,
# flexible load ratio of the total workload
'flexible_load': 0.1,
# Specify reward methods. These are defined in utils/reward_creator.
'ls_reward': 'default_ls_reward',
'dc_reward': 'default_dc_reward',
'bat_reward': 'default_bat_reward',
# Evaluation flag that is required by the load-shifting environment
# To be set only during offline evaluation
'evaluation': False,
# Set this to True if an agent (like MADDPG) returns continuous actions,
"actions_are_logits": False
}