V2G-Predictor

Diploma theses 2020-21

Author: Antonios Valmas
University: NTUA School of electrical and computer engineering
Supervisor: Emmanouel Varvarigos
Description: Optimize the V2G charging system via reinforcement learning and keeping the complexity of the problem independent of the capacity of the parking.

Source code

Requirements

Python 3.8.10 (https://www.python.org/)
pip 21.0.1 (If it isn't installed along with python, check this https://packaging.python.org/tutorials/installing-packages/)
Optional: virtualenv 20.0.28 (https://pypi.org/project/virtualenv/)

Optional: Create virtualenv

Python 3.8.10 must be already installed in your system and accessible through the command line For example, running the following

python3.8 --version

The output should be:

Python 3.8.10

You can create a virtualenv on the root folder of your project by executing the following:

virtualenv -p python3.8 venv

This will create a venv folder containing all files of the python environment

Important: In order to use it you will need to run the following on every command line instance you might need it

source ./venv/bin/activate

Dependencies

All dependencies of the project are documented in the requirements.txt in the root folder fo the project. To install them run the following:

pip install -r requirements.txt

Run

To train the policy

python3 run.py

To evaluate the latest trained policy

python3 eval.py

To export the final plots from the eval data

python3 plot.py

To test the functionalities of the environment

python3 test.py

Abstract modules

The folder app/abstract contains two abstract modules that can be re-used The ddqn module uses the utils module and depends on the folder structure in order to detect it. So if you need to re-use them in a different project under a different project please edit the line 6 on the ddqn.py file and correct the module path

DDQN

The ddqn.py file contains the necessary code to initialize the DDQN agent and train it on a given environment and network. A usage of the module can be found in the app/policies/dqn.py file

Utils

Contains a function compute_avg_return, which computes the average return for a given environment and a given number of episodes to run. Used on the validation step on the train function

wadx2019/V2G-Predictor