CityLearn allows for custom reward function design. The CityLearn Challenge 2022 provides an interface for custom user reward function.
Participants are to edit the get_reward() function in get_reward.py. Three observations from the environment are provided for the reward calculation and they include:
electricity_consumption: List of each building's/total district electricity consumption in [kWh].carbon_emission: List of each building's/total district carbon emissions in [kg_co2].electricity_price: List of each building's/total district electricity price in [$].
By default, the reward function defined in get_reward() is:
$$
\textrm{reward}_n = \textrm{min}(-G_n, 0) + \textrm{min}(-C_n, 0)
$$
Where G_n and C_n are respectively the carbon_emission and electricity_price of the building(s) controlled by agent n.
Note that get_reward() function must return a list whose length is equal to the number of agents in the environment i.e. I the environment uses a single agent, the length of the list is equal to 1 else the length is equal to the number of buildings in the environment.
Do not edit user_reward.py module!