Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Geographically-aware Transformer-based Traffic Forecasting for Urban Motorway Digital Twins

Krešimir Kušić, Vinny Cahill, Ivana Dusparic

TL;DR

This work tackles motorway traffic forecasting for digital twins by introducing GATTF, a Transformer-based model augmented with mutual-information–driven covariates to capture geographically informed spatio-temporal dependencies. MI identifies the most informative sensor covariates, enriching the Transformer’s inputs without increasing model complexity, and yielding substantial accuracy gains on Geneva motorway data. Key findings include significant reductions in MASE and sMAPE for target sensors and clear evidence of non-local sensor interactions driven by network topology. The approach enhances proactive, uncertainty-aware traffic management capabilities and lays groundwork for integration with forecast-driven digital twins in real-world urban networks.

Abstract

The operational effectiveness of digital-twin technology in motorway traffic management depends on the availability of a continuous flow of high-resolution real-time traffic data. To function as a proactive decision-making support layer within traffic management, a digital twin must also incorporate predicted traffic conditions in addition to real-time observations. Due to the spatio-temporal complexity and the time-variant, non-linear nature of traffic dynamics, predicting motorway traffic remains a difficult problem. Sequence-based deep-learning models offer clear advantages over classical machine learning and statistical models in capturing long-range, temporal dependencies in time-series traffic data, yet limitations in forecasting accuracy and model complexity point to the need for further improvements. To improve motorway traffic forecasting, this paper introduces a Geographically-aware Transformer-based Traffic Forecasting GATTF model, which exploits the geographical relationships between distributed sensors using their mutual information (MI). The model has been evaluated using real-time data from the Geneva motorway network in Switzerland and results confirm that incorporating geographical awareness through MI enhances the accuracy of GATTF forecasting compared to a standard Transformer, without increasing model complexity.

Geographically-aware Transformer-based Traffic Forecasting for Urban Motorway Digital Twins

TL;DR

This work tackles motorway traffic forecasting for digital twins by introducing GATTF, a Transformer-based model augmented with mutual-information–driven covariates to capture geographically informed spatio-temporal dependencies. MI identifies the most informative sensor covariates, enriching the Transformer’s inputs without increasing model complexity, and yielding substantial accuracy gains on Geneva motorway data. Key findings include significant reductions in MASE and sMAPE for target sensors and clear evidence of non-local sensor interactions driven by network topology. The approach enhances proactive, uncertainty-aware traffic management capabilities and lays groundwork for integration with forecast-driven digital twins in real-world urban networks.

Abstract

The operational effectiveness of digital-twin technology in motorway traffic management depends on the availability of a continuous flow of high-resolution real-time traffic data. To function as a proactive decision-making support layer within traffic management, a digital twin must also incorporate predicted traffic conditions in addition to real-time observations. Due to the spatio-temporal complexity and the time-variant, non-linear nature of traffic dynamics, predicting motorway traffic remains a difficult problem. Sequence-based deep-learning models offer clear advantages over classical machine learning and statistical models in capturing long-range, temporal dependencies in time-series traffic data, yet limitations in forecasting accuracy and model complexity point to the need for further improvements. To improve motorway traffic forecasting, this paper introduces a Geographically-aware Transformer-based Traffic Forecasting GATTF model, which exploits the geographical relationships between distributed sensors using their mutual information (MI). The model has been evaluated using real-time data from the Geneva motorway network in Switzerland and results confirm that incorporating geographical awareness through MI enhances the accuracy of GATTF forecasting compared to a standard Transformer, without increasing model complexity.
Paper Structure (17 sections, 3 equations, 3 figures, 3 tables)

This paper contains 17 sections, 3 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Mutual Information
  4. GATTF Transformer Architecture for Traffic Forecasting
  5. Transformer architecture
  6. Transformer implementation
  7. Dataset description
  8. Geographically-aware Traffic Forecasting
  9. Spatio-temporal dependencies in traffic flow
  10. Geographically-Aware Covariate Selection
  11. Integrating Spatial Awareness through MI-Based Covariates
  12. MI-Based Selection and Covariate Encoding
  13. RESULTS AND ANALYSIS
  14. Descriptive Analysis
  15. Comparison of traffic prediction performance
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Graph representation of the Geneva motorway showing the most informative detectors for reducing forecasting error of e.g., A6, C3 detectors
  • Figure 2: Geographically-aware input augmentation through MI-based informative covariate selection
  • Figure 3: 24-hour, 5-minute resolution traffic forecast at location C3: (a) GATTF augmented with informative covariates, (b) standard Transformer using all sensors without covariates.