Agents API

Agents API#

The agents in eprllib are defined by its elementary componentes:

  1. The capability of observation.

  2. The action or control habilities.

  3. And the reward, compensation or penalty by its actions given an observation.

_images/agents.png

The image shows the main paramenters inside an agent. All of them must be defined. The AgentSpec class help to construct the agents. To use it just import the class with:

from eprllib.Agent.AgentSpec import AgentSpec

For example, for a simple thermostat definition that take the Zone Mean Air Temperature and put the HVAC on or off:

from eprllib.Agent.AgentSpec import (
    AgentSpec,
    ActionSpec,
    ObservationSpec,
    RewardSpec
)
from eprllib.ActionFunctions.setpoint_control import availability
from eprllib.RewardFunctions.energy_and_cen15251 import reward_fn

HVAC_agent = AgentSpec(
    observation = ObservationSpec(
        variables = [
            ("Zone Mean Air Temperature", "Thermal Zone")
        ]
    ),
    action = ActionSpec(
        actuators = [
            ("Schedule:Constant", "Schedule Value", "HVAC_availability"),
        ],
        action_fn = SetpointAgentActions,
        action_fn_config = {
            'agent_name': 'HVAC_agent',
            'availability_actuator': ("Schedule:Constant", "Schedule Value", "HVAC_availability")
        }
    ),
    reward = RewardSpec(
        reward_fn = reward_fn,
        reward_fn_config = {
            "agent_name": "HVAC_agent",
            "thermal_zone": "Thermal Zone",
            "beta": 0.1,
            'people_name': "Room1 Occupancy",
            'cooling_name': "Cooling:DistrictCooling",
            'heating_name': "Heating:DistrictHeatingWater",
            'cooling_energy_ref': 1500000,
            'heating_energy_ref': 1500000,
    )
)

The agent defined as before is called inside the method EnvConfig.agents() in the argument agents_config as the value for the dictionary.

from eprllib.Env.EnvConfig import EnvConfig

EnvironmentConfig = EnvConfig()
EnvironmentConfig.agents(
    agents_config = {
        'HVAC_agent': HVAC_agent,
    }
)

See the sections of Actions, Observations and Rewards to learn how to configurate each of them.

AgentsConnector API#

A AgentsConnector API is provided to allow agents cooperate or acting in differents ways, like in a hierarchy or in a cooperative manner.

Work in progres…

Contents