Free-floating Electrical Vehicles Sharing Systems (FFEVSS)

This repository contains all implementations of "A Reinforcement Learning Approach for Rebalancing Electric Vehicle Sharing Systems", inclduing environments for both a single and multi shuttle routing in FFEVSS.

Dependencies

Pytorch 1. 5 Python > 3.6

Content

1. agents stores A2CAgents designed to train and test reinforcement learning agents for a single and multi shuttle routing, that use A2C algorithm for training .
2. data containes test dataset designed for a specific urban network structure.
3. envs stores two simulatores to represent environments of FFEVSS with a single and multiple shuttles and data generators that produce training data for simulators on the fly.
4. neuralnets includes Actor and Critic neural nets that are based on sequence to sequence models with attention mechanism
5. results stores inference results based on test datasest stored in the data folder
6. trained_models containes keys of trained models and can be directly loaded to Actor and Critic models
7. main.py imports all needed libraries either for a single or multi shuttle routing and allows to perform training and testing using specific RL agents.
8. network_settings.py specifies an urban network structure of FFEVSS that are later passed to data generator and environment.
9. options.py inlcudes the argument values used throughout training and testing.

To Run

1. Clone this repository.
2. Specify an urban network structure of FFEVSS or use the default structure included in network_settings.py.
3. Run main.py as follows with the default settings to train a single shuttle routing:

python main.py 

4. To perform only inference please set 'train' to False in options.py or just run:

python main.py --train=False

To Cite

@misc{bogyrbayeva2020reinforcement,
      title={A Reinforcement Learning Approach for Rebalancing Electric Vehicle Sharing Systems}, 
      author={Aigerim Bogyrbayeva and Sungwook Jang and Ankit Shah and Young Jae Jang and Changhyun Kwon},
      year={2020},
      eprint={2010.02369},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

Sample Solutions

A sample solution for a sinlge shuttle routing in the urban network of size 23 nodes. A sample solution for a multi shuttle routing in the urban network of size 23 nodes.