Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cross-Modal Reconstruction Pretraining for Ramp Flow Prediction at Highway Interchanges

Yongchao Li, Jun Chen, Zhuoxuan Li, Chao Gao, Yang Li, Chu Zhang, Changyin Dong

TL;DR

Ramp flow prediction at highway interchanges is hindered by missing ramp detectors and data silos. The authors propose STDAE, a two-stage cross-modal pretraining framework that reconstructs ramp flows from mainline data using decoupled spatial and temporal autoencoders, then augments a downstream predictor (GWNet) with learned STDAE representations. Across three real interchanges and multiple sampling intervals, STDAE-GWNET outperforms 13 baselines and approaches the performance of models that have access to historical ramp data, demonstrating robustness to missing inputs and strong generalization. The approach is plug-and-play across predictors and remains effective under data loss, highlighting practical deployment potential for refined highway traffic management. Limitations include lack of weather/incidents data and potential computational costs for real-time network-scale deployment, with future work aimed at broader validation and efficiency improvements.

Abstract

Interchanges are crucial nodes for vehicle transfers between highways, yet the lack of real-time ramp detectors creates blind spots in traffic prediction. To address this, we propose a Spatio-Temporal Decoupled Autoencoder (STDAE), a two-stage framework that leverages cross-modal reconstruction pretraining. In the first stage, STDAE reconstructs historical ramp flows from mainline data, forcing the model to capture intrinsic spatio-temporal relations. Its decoupled architecture with parallel spatial and temporal autoencoders efficiently extracts heterogeneous features. In the prediction stage, the learned representations are integrated with models such as GWNet to enhance accuracy. Experiments on three real-world interchange datasets show that STDAE-GWNET consistently outperforms thirteen state-of-the-art baselines and achieves performance comparable to models using historical ramp data. This demonstrates its effectiveness in overcoming detector scarcity and its plug-and-play potential for diverse forecasting pipelines.

Cross-Modal Reconstruction Pretraining for Ramp Flow Prediction at Highway Interchanges

TL;DR

Ramp flow prediction at highway interchanges is hindered by missing ramp detectors and data silos. The authors propose STDAE, a two-stage cross-modal pretraining framework that reconstructs ramp flows from mainline data using decoupled spatial and temporal autoencoders, then augments a downstream predictor (GWNet) with learned STDAE representations. Across three real interchanges and multiple sampling intervals, STDAE-GWNET outperforms 13 baselines and approaches the performance of models that have access to historical ramp data, demonstrating robustness to missing inputs and strong generalization. The approach is plug-and-play across predictors and remains effective under data loss, highlighting practical deployment potential for refined highway traffic management. Limitations include lack of weather/incidents data and potential computational costs for real-time network-scale deployment, with future work aimed at broader validation and efficiency improvements.

Abstract

Interchanges are crucial nodes for vehicle transfers between highways, yet the lack of real-time ramp detectors creates blind spots in traffic prediction. To address this, we propose a Spatio-Temporal Decoupled Autoencoder (STDAE), a two-stage framework that leverages cross-modal reconstruction pretraining. In the first stage, STDAE reconstructs historical ramp flows from mainline data, forcing the model to capture intrinsic spatio-temporal relations. Its decoupled architecture with parallel spatial and temporal autoencoders efficiently extracts heterogeneous features. In the prediction stage, the learned representations are integrated with models such as GWNet to enhance accuracy. Experiments on three real-world interchange datasets show that STDAE-GWNET consistently outperforms thirteen state-of-the-art baselines and achieves performance comparable to models using historical ramp data. This demonstrates its effectiveness in overcoming detector scarcity and its plug-and-play potential for diverse forecasting pipelines.

Paper Structure

This paper contains 22 sections, 22 equations, 7 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Short-term traffic prediction
  4. Ramp flow prediction
  5. Pre-training framework for traffic prediction
  6. Methodology
  7. Problem Statement
  8. The pre-training model
  9. Input Feature Encoding
  10. STDAE
  11. Downstream prediction model
  12. Dataset and processing
  13. Define the scope
  14. Data collection
  15. Data cleaning
  16. ...and 7 more sections

Figures (7)

  • Figure 1: Illustration of the three primary challenges faced by ramp flow prediction based on mainline data. (a) Data silos: vehicle records are fragmented among different highway operators, preventing cross-regional vehicle tracking. (b) Data loss: sensor failures, adverse weather, or transmission errors can cause missing values even in mainline data. (c) Real-time blind spot: real-time ramp flow data are unavailable due to privacy restrictions and system integration barriers, although historical data can be used for model training.
  • Figure 2: STDAE Pre-training and Prediction Framework
  • Figure 3: Data processing workflow, including the determination of the data collection scope, data collection, data cleaning, and feature extraction.
  • Figure 4: Comparison of MAPE across different sampling intervals. (a) Average MAPE of 14 models. (b) MAPE of the STDAEGWNET model.
  • Figure 5: Comparison of STDAE and its ablated versions on three datasets with 5 min sampling interval.
  • ...and 2 more figures