This is the official code of STKEC, "Knowledge Expansion and Consolidation for Continual Traffic Prediction With Expanding Graphs", which is a contintual traffic prediction framework and published in the journal TITS2023. This code is based on TrafficStream.

Please download the data from TrafficStream(https://github.com/AprLie/TrafficStream) and follow the same processing method.

conda env create -f STKEC.yaml

python main.py --conf STKEC.json --gpuid 4

#TODO: if you want to use another dataset, please aggregate the day vector of each node, and complete long_pattern=np.load('long_term_path') in line 76 in main.py.

#ToDO: select some nodes with stable patterns to evaluate the effectiveness of knowledge consolidation and select new nodes to to evaluate the effectiveness of knowledge expand.

Please cite our paper if you find it useful:

@article{wang2023knowledge, title={Knowledge Expansion and Consolidation for Continual Traffic Prediction With Expanding Graphs}, author={Wang, Binwu and Zhang, Yudong and Shi, Jiahao and Wang, Pengkun and Wang, Xu and Bai, Lei and Wang, Yang}, journal={IEEE Transactions on Intelligent Transportation Systems}, year={2023}, publisher={IEEE} }