TemperatureControl-ReinforcementLearning

Comparing the performance of a DDPG Reinforcement learning model to control temperature with that of a PID and a thermostat controller.

Find video of the training process here.

Find the Google Slides Link to the project presentation here.

Thermostat controller controlling the temperature.

Mean Square Error - 32.7782

PID controller controlling the temperature.

Mean Square Error - 23.9247

DDPG RL Agent controller controlling the temperature.

Mean Square Error - 26.8667

Steps to recreate models:

1. Run sldemo_househeat_data.m, and make sure variables exist on the workspace.
2. Run house_thermostat.slx to generate a the plots for the control using a regular thermostat.
3. Run house_PID.slx to generate a the plots for the control using a Discrete PID controller.
4. Open the ddpg_live(new).mlx live notebook. Start running each cell individually. (Make sure the variable - training is set to true in the notebook.)

Warnings:

Make sure to have the following toolkits installed to be able to recreate these simulations successfully:

1. Reinforcement Learning Toolkit.
2. Machine Learning Toolkit.
3. PID Tuner

Run files using MATLAB 2020.

Create your own reward function to tune model results.

You will be able to tune the reward function for the simulation by updating the Reward block in the RL_Heat_DDPG_test.slx file.

Use this link to set up base thermal model of the house from the MATLAB-SIMULINK website.