Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SGRU: A High-Performance Structured Gated Recurrent Unit for Traffic Flow Prediction

Wenfeng Zhang, Xin Li, Anqi Li, Xiaoting Huang, Ti Wang, Honglei Gao

TL;DR

This paper addresses traffic-flow prediction by reframing it as a multivariate time-series problem and identifying limitations of dilated convolutions and time-slice methods in capturing adjacent-step and inter-slice dependencies. It introduces Structured Gated Recurrent Units (SGRU) that fuse adaptive graph convolution, multilayer spatiotemporal embedding, and a novel Connection Module to fuse parallel GRU outputs, enabling effective modeling of full sequence dynamics. Empirical results on PeMS03/04/07/08 show substantial improvements (average MAE/RMSE/MAPE gains around 11-19%) over diverse baselines, with ablations confirming the contributions of the embedding and structured GRU components. The approach offers practical benefits for accurate traffic forecasting in smart cities, and the authors provide publicly accessible code for reproducibility.

Abstract

Traffic flow prediction is an essential task in constructing smart cities and is a typical Multivariate Time Series (MTS) Problem. Recent research has abandoned Gated Recurrent Units (GRU) and utilized dilated convolutions or temporal slicing for feature extraction, and they have the following drawbacks: (1) Dilated convolutions fail to capture the features of adjacent time steps, resulting in the loss of crucial transitional data. (2) The connections within the same temporal slice are strong, while the connections between different temporal slices are too loose. In light of these limitations, we emphasize the importance of analyzing a complete time series repeatedly and the crucial role of GRU in MTS. Therefore, we propose SGRU: Structured Gated Recurrent Units, which involve structured GRU layers and non-linear units, along with multiple layers of time embedding to enhance the model's fitting performance. We evaluate our approach on four publicly available California traffic datasets: PeMS03, PeMS04, PeMS07, and PeMS08 for regression prediction. Experimental results demonstrate that our model outperforms baseline models with average improvements of 11.7%, 18.6%, 18.5%, and 12.0% respectively.

SGRU: A High-Performance Structured Gated Recurrent Unit for Traffic Flow Prediction

TL;DR

This paper addresses traffic-flow prediction by reframing it as a multivariate time-series problem and identifying limitations of dilated convolutions and time-slice methods in capturing adjacent-step and inter-slice dependencies. It introduces Structured Gated Recurrent Units (SGRU) that fuse adaptive graph convolution, multilayer spatiotemporal embedding, and a novel Connection Module to fuse parallel GRU outputs, enabling effective modeling of full sequence dynamics. Empirical results on PeMS03/04/07/08 show substantial improvements (average MAE/RMSE/MAPE gains around 11-19%) over diverse baselines, with ablations confirming the contributions of the embedding and structured GRU components. The approach offers practical benefits for accurate traffic forecasting in smart cities, and the authors provide publicly accessible code for reproducibility.

Abstract

Traffic flow prediction is an essential task in constructing smart cities and is a typical Multivariate Time Series (MTS) Problem. Recent research has abandoned Gated Recurrent Units (GRU) and utilized dilated convolutions or temporal slicing for feature extraction, and they have the following drawbacks: (1) Dilated convolutions fail to capture the features of adjacent time steps, resulting in the loss of crucial transitional data. (2) The connections within the same temporal slice are strong, while the connections between different temporal slices are too loose. In light of these limitations, we emphasize the importance of analyzing a complete time series repeatedly and the crucial role of GRU in MTS. Therefore, we propose SGRU: Structured Gated Recurrent Units, which involve structured GRU layers and non-linear units, along with multiple layers of time embedding to enhance the model's fitting performance. We evaluate our approach on four publicly available California traffic datasets: PeMS03, PeMS04, PeMS07, and PeMS08 for regression prediction. Experimental results demonstrate that our model outperforms baseline models with average improvements of 11.7%, 18.6%, 18.5%, and 12.0% respectively.
Paper Structure (21 sections, 9 equations, 6 figures, 2 tables)

This paper contains 21 sections, 9 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Problem 1: Dilated convolutions break adjacent time steps
  3. Problem 2: Time slices cause a weak of spatial connectivity
  4. Our solution
  5. Related Works
  6. Static Adjacency Matrices
  7. Dynamic Adjacency Matrices
  8. Preliminary
  9. Methodology
  10. Preparation & Embedding
  11. Adaptive graph convolution.
  12. Multilayer spatiotemporal embedding.
  13. Temporal feature extraction.
  14. Structured GRUs network.
  15. Prediction and loss calculation.
  16. ...and 6 more sections

Figures (6)

  • Figure 1: Traffic Flow Prediction: a typical MTS problem
  • Figure 2: Traffic waveforms of two adjacent nodes (they are similar with a lag of about 30 minutes)
  • Figure 3: Model structure of SGRU
  • Figure 4: Structural comparison between SGRU (left) and Linear GRUs (right)
  • Figure 5: PeMS03 visualization of nodes No.10 and No.20
  • ...and 1 more figures