STDEN: Towards Physics-Guided Neural Networks for Traffic Flow Prediction

Authors: Jiahao Ji, Jingyuan Wang, Zhe Jiang, Jiawei Jiang, Hu Zhang4048-4056

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on three real-world traffic datasets in Beijing show that our model outperforms state-of-the-art baselines by a significant margin. A case study further verifies that STDEN can capture the mechanism of urban traffic and generate accurate predictions with physical meaning. We conduct extensive experiments on three real-world traffic datasets, and the proposed method achieves significant improvement over state-of-the-art baselines.
Researcher Affiliation Academia Jiahao Ji, 1 Jingyuan Wang, 1,2* Zhe Jiang, 3 Jiawei Jiang, 4 Hu Zhang 4 1 State Key Laboratory of Software Development Environment, School of Computer Science & Engineering, Beihang University, Beijing, China 2 Peng Cheng Laboratory, Shenzhen, China 3 Department of Computer & Information Science & Engineering, The University of Florida 4 MOE Engineering Research Center of Advanced Computer Application Technology, School of Computer Science & Engineering, Beihang University, Beijing, China
Pseudocode No The paper includes mathematical equations and network architecture diagrams but does not contain any explicit pseudocode blocks or algorithms.
Open Source Code Yes The code is available at https://github.com/Echo-Ji/STDEN
Open Datasets No We evaluate the performance of our model over the real-world urban traffic dataset collected by the Beijing Municipal Commission of Transport, which contains trajectories of 40,000 taxies in Beijing from April 1st 2015 to July 31st 2015 (totally 4 months). The paper describes how the dataset was collected but does not provide any link, DOI, or formal citation for public access.
Dataset Splits Yes We split each dataset into the training, validation, and test sets with a ratio of 7:1:2.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU/CPU models, memory specifications).
Software Dependencies No The paper mentions software components like 'dopri5' and 'GRU' but does not specify their version numbers or other library dependencies with versions.
Experiment Setup Yes The settings of STDEN contains the following two parts: (1) Settings of the DE-Net part. We model the dynamics of potential energy in latent space using an adaptive method dopri5 (Dormand and Prince 1980), and conduct grid search on the latent dimension over {1, 2, 4, 8}. (2) Settings of the encoder. GRU is used to encode the distribution of the initial value of the PEF-sequence. The number of hidden units in GRU is searched over {16, 32, 64, 128}. More implementation details about our STDEN and other baselines settings are given in Appendix.