CODEWITHSUNDEEP
machine-learningtensorflowartificial-intelligencereinforcement-learning
Leb_Broth
Leb_Broth

Reputation: 1111

Q-learning with 2D actions and 2D states

My problem is as follows:

The agent should at each state, adjust the water flow and a fan speed for a power plant boiler to receive a feedback of a double state: current temperature, amount of emissions.

If my agent has a tuple of actions and a tuple of states, does that mean i should split my q-learning problem into 2 where 1 agent would have a Q and R matrix for the water/temperature environment and the other agent for the fan speed/amount of emissions environment? Or is there a way to represent an R and Q matrix for the agent described originally?

Upvotes: 0

Views: 1057

Answers (1)

Don Reba
Don Reba

Reputation: 14031

It is normal for states and actions to be multidimensional. What you do is have your agent learn the values of all combinations of water flow and fan speed for all combinations of current temperature and amount of emission. If this makes the table unwieldy, then you will need to approximate it, and this is a whole field in itself.

Upvotes: 1

Related Questions