Irregular Traffic Time Series Forecasting Based on Asynchronous Spatio-Temporal Graph Convolutional Network

TL;DR

This work tackles irregular traffic forecasting produced by adaptive traffic signal controls, where sensor measurements are asynchronous and temporal gaps vary. It introduces ASeer, an integrated framework combining an Asynchronous Graph Diffusion Network for spatial diffusion, a Transformable Time-aware Convolution Network with Personalized Time Encoding for irregular temporal dynamics, and a Semi-Autoregressive Prediction Network for efficient variable-length forecasting. The model is trained with masked losses to handle missing data and validated on two real-world Chinese city datasets, where it achieves state-of-the-art accuracy across six metrics. The findings demonstrate that explicitly modeling asynchronous spatial dependencies and irregular temporal patterns yields substantial gains for intelligent transportation systems and real-time urban traffic management.

Abstract

Accurate traffic forecasting is crucial for the development of Intelligent Transportation Systems (ITS), playing a pivotal role in modern urban traffic management. Traditional forecasting methods, however, struggle with the irregular traffic time series resulting from adaptive traffic signal controls, presenting challenges in asynchronous spatial dependency, irregular temporal dependency, and predicting variable-length sequences. To this end, we propose an Asynchronous Spatio-tEmporal graph convolutional nEtwoRk (ASeer) tailored for irregular traffic time series forecasting. Specifically, we first propose an Asynchronous Graph Diffusion Network to capture the spatial dependency between asynchronously measured traffic states regulated by adaptive traffic signals. After that, to capture the temporal dependency within irregular traffic state sequences, a personalized time encoding is devised to embed the continuous time signals. Then, we propose a Transformable Time-aware Convolution Network, which adapts meta-filters for time-aware convolution on the sequences with inconsistent temporal flow. Additionally, a Semi-Autoregressive Prediction Network, comprising a state evolution unit and a semi-autoregressive predictor, is designed to predict variable-length traffic sequences effectively and efficiently. Extensive experiments on a newly established benchmark demonstrate the superiority of ASeer compared with twelve competitive baselines across six metrics.

Figures (11)

• Figure 1: The distinction between classical traffic forecasting on the highway and irregular traffic forecasting under adaptive traffic signal control. The forecasting task aims to predict future traffic variations (10:00-11:00) based on past observations (9:00-10:00).
• Figure 2: The framework overview of ASeer, which consists of three major components: Asynchronous Graph Diffusion Network (AGDN), Transformable Time-aware Convolution Network (TTCN), and Semi-Autoregressive Prediction Network (SAPN). The traffic states are first inputted to AGDN to obtain spatial representations, which are incorporated by TTCN to acquire the spatiotemporal representations. After that, SAPN predicts the variable-length traffic state sequence based on the spatiotemporal representations. Throughout the entire process, personalized time encoding is used to embed continuous time.
• Figure 3: Results of ablation study. "Z" and "B" denote Zhuzhou and Baoding, respectively.
• Figure 4: Effect of different prediction step sizes.
• Figure 5: Effect of different hidden dimensions.