Updates

• ST-EVCDP-Version2 is coming soon. The data will span from September 2022 to September 2023, with granularity to charging stations, including coordinates, numbers of chargers, occupancy, and price. Other information will also be gradually released after being desensitized for academic research purposes.
• May 12, 2024: We uploaded the data of weather conditions in the studied areas, namely SZweather20220619-20220718.csv and SZweather_Header.txt.
• March 15, 2024: We uploaded the data of charging duration and volume in the studied areas.

Spatio-temporal EVCDP (Shenzhen)

This is a real-world dataset for spatio-temporal electric vehicle (EV) charging demand prediction. If it is helpful to your research, please cite our paper:

Qu, H., Kuang, H., Li, J., & You, L. (2023). A physics-informed and attention-based graph learning approach for regional electric vehicle charging demand prediction. arXiv preprint arXiv:2309.05259. Paper in arXiv

Author: Haohao Qu (haohao.qu@connect.polyu.hk)

Data Description

The data used in this study is drawn from a publicly available mobile application, which provides the real-time availability of charging piles (i.e., idle or not). Within Shenzhen, China, a total of 18,061 public charging piles are covered during the studied period from 19 June to 18 July 2022 (30 days) with a minimum interval of 5 minutes and 8640 timestamps. As shown in Figure 1, the city is constructed into a graph-structure data with 247 nodes (traffic zones) and 1006 edges (adjacent relationships).

Figure 1. Spatial distribution of public EV charging piles in Shenzhen.

Besides, the pricing schemes for the studied charging piles are also collected. Among the 247 traffic zones, 57 of them (enclosed in red lines) deploy time-based pricing schemes, while others use fixed ones. More statistical details are illustrated in the following table.

Files

• adj.csv: The adjacent matrix of studied areas, 1 indicates the two traffic zones are neighboring, vice versa.
• distance.csv: Distances between nodes.
• information.csv: Several basis information about the data, including pile capacity, longitude, latitude, whether or not located in the central business district (1:yes, 0:no), and whether or not on a time-based pricing scheme (1:yes, 0:no).
• occupancy.csv: The real-time EV charging occupancy in studied areas.
• duration.csv: The real-time EV charging duration in studied areas, i.e., the sum of charging time for all charging piles, unit in hour.
• volume.csv: The real-time EV charging volume in studied areas, i.e., the total power consumption of all charging piles, unit in kWh.
• price.csv: The real-time EV charging pricing in studied areas.
• time.csv: The timestamps of studied period.
• Shenzhen.qgz: The QGIS map file of Shenzhen city.

Enviroment Requirement

pip install -r requirements.txt

An simple example to run Spatio-temporal Prediction on the dataset

We developed a physics-informed and attention-based approach for spatio-temporal EV charging demand prediction, named PAG. Expect that, some representative methods are included, e.g., LSTM, and GCN-LSTM, GAT-LSTM. You can train and test the proposed model through the following procedures:

1. Choose your model in line 45 of main.py or use the default model (PAG-) by skipping this procedure.
2. Run main.py via Pycharm, etc. or change your ROOT_PATH and command:

cd [path] && python main.py

Extend:

• If you want to run your own models on the datasets we offer, you should go to models.py and replace the model in main.py.

More updates will be posed in the near future! Thank you for your interest.