https://github.com/ntt-dkiku/evrp-eps
The official implementation of "Electric Vehicle Routing for Emergency Power Supply with Deep Reinforcement Learning" (AAMAS 2024, extended abstract).
https://github.com/ntt-dkiku/evrp-eps
combinatorial-optimization deep-reinforcement-learning electric-vehicle-routing-problem electric-vehicles multi-agent-reinforcement-learning multi-agent-systems transformer vehicle-routing-problem
Last synced: 4 months ago
JSON representation
The official implementation of "Electric Vehicle Routing for Emergency Power Supply with Deep Reinforcement Learning" (AAMAS 2024, extended abstract).
- Host: GitHub
- URL: https://github.com/ntt-dkiku/evrp-eps
- Owner: ntt-dkiku
- License: other
- Created: 2024-01-09T08:07:07.000Z (almost 2 years ago)
- Default Branch: main
- Last Pushed: 2024-04-08T02:49:07.000Z (over 1 year ago)
- Last Synced: 2025-02-15T11:39:05.766Z (8 months ago)
- Topics: combinatorial-optimization, deep-reinforcement-learning, electric-vehicle-routing-problem, electric-vehicles, multi-agent-reinforcement-learning, multi-agent-systems, transformer, vehicle-routing-problem
- Language: Jupyter Notebook
- Homepage: https://ntt-dkiku.github.io/rl-evrpeps/
- Size: 24 MB
- Stars: 4
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Electric Vehicle Routing for Emergency Power Supply: Towards Telecom Base Station Relief
This repo is the official implementation of Electric Vehicle Routing for Emergency Power Supply with Deep Reinforcement Learning (AAMAS 2024, extended abstract) and [Electric Vehicle Routing for Emergency Power Supply: Towards Telecom Base Station Relief](https://arxiv.org/abs/2404.02448) (arXiv preprint).
## 📦 Python Environment
We recommend using Docker to construct the python environment. You can use the [Dockerfile](./Dockerfile) in this repository.
```
docker build -t evrp-eps/evrp-eps:1.0 .
```
You can run code interactively with the following command (<> indicates a placeholder, which you should replace according to your settings.").
```
docker run -it --rm -v :/workspace/app --name evrp-eps -p : --gpus all evrp-eps/evrp-eps:1.0 bash
```
## 🔧 Usage
### 1. Generating synthetic data
First of all, we generate synthetic datasets for training/validation/evaluation. We recommend more than 12.8M training samples.
If you want to change some parameters, check the other options with ```python generate_datasets.py -h```.
```
python generate_dataset.py --save_dir data/synthetic_data --type all --num_samples 1280000 10000 10000
```
### 2. Training
We train the RL model on the synthetic datasets. Check the other options with ```python train.py -h```.
```
python train.py --dataset_path data/synthetic_data/train_dataset.pkl --checkpoint_dir checkpoints/demo_model --batch_size 256 --vehicle_speed 41 --wait_time 0.5 --time_horizon 12 --gpu 0
```
### 3. Validation
We determine the best epoch evaluating the RL-model with greedy decoding on the validation split. Check the other options with ```python valid.py -h```.
```
python valid.py --model_dir checkpoints/demo_model --dataset_path data/synthetic_data/valid_dataset.pkl --gpu 0
```
### 4. Evaluation
The option ```--model_dir ``` automatically selects the weights at the best epoch. You can also select a specific epoch by ```--model_path model_epoch.pth```. Specify only one of the two. If you want to output the visualization, add the option ```--visualize_routes```. Check the other options with ```python eval.py -h```.
```
python eval.py --model_dir checkpoints/demo_model --dataset_path data/synthetic_data/eval_dataset.pkl --vehicle_speed 41 --wait_time 0.5 --time_horizon 12 --gpu 0
```
## 🧪 Reproducibility
Regarding the synthetic datasets, you can reproduce our experimental results in [reproduce_results.ipynb](./reproduce_results.ipynb).
Please take a glance at the content via nbviwer: [nbviwer-evrp-eps](https://nbviewer.org/github/ntt-dkiku/evrp-eps/blob/main/reproduce_results.ipynb).
You can open the Jupyter Notebook server with the following command inside the container, then access it from your browser on localhost.
```
jupyter lab --allow-root --no-browser --ip=0.0.0.0 --port
```
## 🐞 Bug reports and questions
If you encounter a bug or have any questions, please post issues in this repo.
## 📄 Licence
Our code is licenced by NTT. Basically, the use of our code is limitted to research purposes. See [LICENSE](./LICENSE) for more details.
## 🤝 Citation
If you find this work useful, please cite our paper as follows:
```
@misc{kikuta2024electric,
title={Electric Vehicle Routing Problem for Emergency Power Supply: Towards Telecom Base Station Relief},
author={Daisuke Kikuta and Hiroki Ikeuchi and Kengo Tajiri and Yuta Toyama and Masaki Nakamura and Yuusuke Nakano},
year={2024},
eprint={2404.02448},
archivePrefix={arXiv},
primaryClass={math.OC}
}
```