Low-Rank and Deep Plug-and-play Priors for Missing Traffic Data Imputation

This is the code repository for our works on low-rank tensor completion for missing traffic data imputation:

[1] Peng Chen, Fang Li, Deliang Wei, and Changhong Lu, "Spatiotemporal traffic data completion with truncated minimax-concave penalty," Transportation Research Part C: Emerging Technologies, vol. 164, July 2024, Art. no. 104657.

[2] Peng Chen, Fang Li, Deliang Wei, and Changhong Lu, "Low-Rank and Deep Plug-and-Play Priors for Missing Traffic Data Imputation," IEEE Transactions on Intelligent Transportation Systems (Early Access), pp. 1-17, Nov. 2024.

Methodology

The general procedure of our proposed methods is as follows: Initially, we conduct each mode unfolding matrix of $\mathcal{M}^t$ to low-rank processing, supplemented by optional deep PnP processing; Subsequently, upon folding and weighted aggregation of the processed matrices, we correlate the resultant $\widehat{\mathcal{M}}^t$ with the observation tensor $\mathcal{Y}$ to derive the tensor $\mathcal{M}^{t+1}$ for the subsequent iteration.

Preparation

Data preparation

We provide four selected public traffic datasets in the shared folder Google Drive. More related datasets can refer to the transdim project. Please download the datasets and put them in the ./datasets folder. The overview of the provided datasets is as follows:

• Hangzhou: Hangzhou metro passenger flow dataset. This dataset contains information on incoming passenger flow for 80 metro stations in Hangzhou, China. The data covers a period of 25 days, from January 1st to January 25th, 2019, with a 10-minute resolution. The time interval from 0:00 a.m. to 6:00 a.m., when there are no services, has been excluded. Only the remaining 108 time intervals of a day are considered. The dataset is presented as a tensor of size $'80 \times 25 \times 108'$ ($'80 \times 2700'$ in the form of a time series matrix).

• Portland: Portland highway traffic volume dataset. This dataset comprises traffic volume data collected from highways in the Portland-Vancouver Metropolitan region in January 2021. It was obtained from 1156 loop detectors with a 15-minute resolution, resulting in 96 time intervals per day. The dataset is in the form of a tensor of size $'1156 \times 31 \times 96'$ ($'1156 \times 2976'$ in the form of a time series matrix).

• Seattle: Seattle freeway traffic speed dataset. This dataset contains information on the speed of freeway traffic in Seattle, USA. The data was collected from 323 loop detectors with a 5-minute resolution, resulting in 288 time intervals per day. The data is presented as a tensor of size $'323 \times 28 \times 288'$ ($'323 \times 8064'$ when presented as a time series matrix).

• PeMS: PeMS freeway traffic volume dataset. This dataset includes the traffic volume recorded by 228 loop detectors in District 7 of California, with a 5-minute time resolution. The data was collected over the weekdays of May and June in 2012 by Caltrans Performance Measurement System (PeMS). The data is in the form of a tensor of size $'228 \times 44 \times 288'$ ($'228 \times 12672'$ in the form of the time series matrix).

Besides, we provide the pretrained parameter of the used light DRUNet in Google Drive for the deep plug-and-play prior. Please download the parameter and put it in the ./drunet_light_params folder.

Environment preparation

The required packages are listed in the requirements.txt file. You can run the following shell command to create a new environment named lrtc and install the packages by running the following command:

conda create --name lrtc --file requirements.txt

The default command installs the CPU version of PyTorch. For faster computation of deep PnP processing using NVIDIA's CUDA, install the CUDA version based on your hardware and system. Check the official website for installation instructions.

Cite Information

If you find this repo useful for your research, please consider citing the papers:

@article{CHEN2024104657,
    title = {Spatiotemporal traffic data completion with truncated minimax-concave penalty},
    journal = {Transportation Research Part C: Emerging Technologies},
    volume = {164},
    pages = {104657},
    year = {2024},
    issn = {0968-090X},
    doi = {https://doi.org/10.1016/j.trc.2024.104657},
    url = {https://www.sciencedirect.com/science/article/pii/S0968090X24001785},
    author = {Peng Chen and Fang Li and Deliang Wei and Changhong Lu},
    publisher={Elsevier}
}

@ARTICLE{10756233,
  author={Chen, Peng and Li, Fang and Wei, Deliang and Lu, Changhong},
  journal={IEEE Transactions on Intelligent Transportation Systems}, 
  title={Low-Rank and Deep Plug-and-Play Priors for Missing Traffic Data Imputation}, 
  year={2024},
  volume={},
  number={},
  pages={1-17},
  doi={https://doi.org/10.1109/TITS.2024.3493864}
}