Code used for the paper 'Lithium-Ion Battery Management System with Reinforcement Learning for Balancing State of Charge and Cell Temperature'. Please read the paper for further information.
A conda environment with all necessary dependencies can be set up with environment.yml.
The folder batterymodel contains cell.py for the implementation of a cell and battery.py for a battery pack with multiple cells. The implementation follows the modified Shepherd model. The file battery.py already includes code necessary for the simulated usage of an active battery management system.
The reinforcement learning environment is defined in the file env.py following the gym interface of stable baselines 3.
The code fore the hyperparameter optimization with HyperOpt is defined in optimize_parallel.py. Settings for the training can be made via config.yml.
To execute the hyperparameter optimization in parallel the following needs to be done (with the corresponding conda environment activated in all terminals):
-
Start the mongod server in a terminal:
mongod --dbpath . --port 1234 --directoryperdb --journal -
Open a new terminal to run the hyperparameter optimization in with
from optimize_parallel import optimize_ppo optimize_ppo()
The program stops without executing any trials. To run trials stept 3 is necessary.
-
For an execution of n trials in parallel repeat the following n times: start a new terminal and run
PYTHONPATH=. hyperopt-mongo-worker --mongo=localhost:1234/trial_db --poll-interval=0.1