Document Type

Thesis - Open Access

Award Date

2025

Degree Name

Master of Science (MS)

Department / School

Electrical Engineering and Computer Science

First Advisor

Jun Huang

Abstract

Accurate short-term traffic forecasting is central to modern Intelligent Transportation Systems, supporting route guidance, adaptive signal control, and incident response. Yet producing reliable predictions remains difficult because traffic is highly non-stationary. The relationships among roadway sensors shift during congestion, incidents, weather changes, or fluctuations in demand, and the temporal structure of traffic spans several scales from abrupt minute-level variations to broader daily and weekly rhythms. Models that rely on fixed spatial graphs or a single temporal scale tend to miss these evolving and layered dependencies. This thesis addresses these challenges by developing a graph-learning framework that adapts to changing traffic conditions while remaining computationally efficient and interpretable. The proposed model, ST-Hybrid, learns spatial connectivity dynamically based on the current traffic state. A lightweight graph learner combines adaptive node embeddings with observed features to construct a state-dependent adjacency matrix at each forecasting step. This dynamic graph is sparsified for efficiency and fused with the underlying road topology. To capture temporal structure, the model processes multiple scales of variation through parallel dilated convolutions, and applies multi-head spatial attention to identify which sensors are most influential for each prediction, providing transparency into the forecasting process. To further ground the learning process in physical realism, the framework is extended with two principles from traffic flow theory: flow conservation, which preserves vehicle counts across connected segments, and backward-moving wave propagation, which models how congestion travels upstream. These cues help regularize the adaptive graph so that its structure remains consistent with known traffic behavior. A temporal causality constraint is also introduced to ensure that predictions depend only on information from valid prior states, improving the stability and coherence of the model. The approach is evaluated on the standard Caltrans PeMSD3, PeMSD4, PeMSD7 and PeMSD8 datasets. ST-Hybrid achieves strong performance across all four benchmark networks. On PeMSD8, it obtains a mean absolute error (MAE) of 15.19 and root mean square error (RMSE) of 24.18, ranking second among recent dynamic-graph approaches. It maintains competitive accuracy on the larger PeMSD4 network (MAE 20.16, RMSE 32.17) and demonstrates robust performance on PeMSD3 (MAE 15.58, RMSE 25.93) and PeMSD7 (MAE 21.58, RMSE 34.36). Analysis reveals consistent patterns across the datasets. The dynamic graph component provides the most substantial benefit in networks with higher volatility and frequently changing connectivity, while the multi-scale temporal module delivers reliable improvements across all forecasting horizons from 5 to 60 minutes. A detailed node-level breakdown shows that the majority of residual errors originate from a small subset of unstable sensors, suggesting that future refinements can productively focus on these specific outliers rather than implementing broad structural changes to the model architecture. Importantly, ST-Hybrid maintains low inference latency throughout these evaluations, demonstrating its suitability for real-time, city-scale deployment in practical transportation management systems. Overall, this work demonstrates that traffic forecasting models can successfully adapt spatial connectivity to real-time conditions, capture complex multi-scale temporal dynamics, and provide interpretable, operationally useful insights. The framework offers a practical and reliable foundation for real-world transportation forecasting and decision-making.

Library of Congress Subject Headings

Traffic estimation. 
Graph theory. 
Machine learning -- Graphic methods.      

Publisher

South Dakota State University

Download

Share

COinS
 

Rights Statement

In Copyright