This is the official implementation of KDD 24 paper Irregular Traffic Time Series Forecasting Based on Asynchronous Spatio-Temporal Graph Convolutional Networks.

New Benchmark and Datasets

We meticulously collect and develop two novel real-world datasets of irregular traffic time series, and establish a systematic evaluation scheme comprising six metrics, setting a new benchmark in the field of irregular traffic forecasting.

Datasets are available at the provided dataset link, where the files for each dataset represents:

dataset_onhour.pkl: (x, y, mask_x, mask_y)

• x fields: end time of cycle, cycle length, traffic flow, unit traffic flow
  
• y fields: begin time of cycle, cycle length, traffic flow, unit traffic flow
  
• mask_x: mask terms of x
  
• mask_y: mask terms of y
  

distance_geo.npy: pair-wise geographical distance between sensors.
reachability.npy: pair-wise lane-level road network reachability between sensors.

More detailed analysis on the datasets can be found in the paper Appendix A.1 Data Description and Analysis.

The main results:

Requirements

ASeer has been tested using Python 3.8.8.

To have consistent libraries and their versions, you can install the needed dependencies for this project by running the following command:

pip install -r requirements.txt

Run the Model

Take dataset ZHUZHOU as an example.

You need to first execute the below command to generate the asynchronously diffused message edges for the dataset:

python ./lib/message_edges_gen.py zhuzhou

To replicate the experimental results presented in the paper, please execute the below script:

sh ./scripts/zhuzhou.sh

Citation

@inproceedings{10.1145/3637528.3671665,
author = {Zhang, Weijia and Zhang, Le and Han, Jindong and Liu, Hao and Fu, Yanjie and Zhou, Jingbo and Mei, Yu and Xiong, Hui},
title = {Irregular Traffic Time Series Forecasting Based on Asynchronous Spatio-Temporal Graph Convolutional Networks},
year = {2024},
booktitle = {Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
pages = {4302–4313}
}