Spatio-Temporal Graph Convolutional Networks: A Deep Learning Framework for Traffic Forecasting

TL;DR

This paper tackles the challenge of accurate mid- and long-term traffic forecasting by modeling traffic data as graph signals and introducing STGCN, a fully convolutional framework that couples graph convolutions with gated temporal convolutions. STGCN uses spatio-temporal blocks (ST-Conv) to efficiently extract spatial dependencies from graph topology and temporal dynamics from sequential data, aided by Chebyshev polynomial and first-order approximations for scalable graph filtering. The approach achieves superior predictive accuracy and training efficiency over strong baselines on BJER4 and PeMSD7 datasets, while using fewer parameters. The results demonstrate STGCN's potential for large-scale, real-time traffic forecasting and its applicability to other spatio-temporal graph-structured domains.

Abstract

Timely accurate traffic forecast is crucial for urban traffic control and guidance. Due to the high nonlinearity and complexity of traffic flow, traditional methods cannot satisfy the requirements of mid-and-long term prediction tasks and often neglect spatial and temporal dependencies. In this paper, we propose a novel deep learning framework, Spatio-Temporal Graph Convolutional Networks (STGCN), to tackle the time series prediction problem in traffic domain. Instead of applying regular convolutional and recurrent units, we formulate the problem on graphs and build the model with complete convolutional structures, which enable much faster training speed with fewer parameters. Experiments show that our model STGCN effectively captures comprehensive spatio-temporal correlations through modeling multi-scale traffic networks and consistently outperforms state-of-the-art baselines on various real-world traffic datasets.

Figures (5)

• Figure 1: Graph-structured traffic data. Each $v_t$ indicates a frame of current traffic status at time step $t$, which is recorded in a graph-structured data matrix.
• Figure 2: Architecture of spatio-temporal graph convolutional networks. The framework STGCN consists of two spatio-temporal convolutional blocks (ST-Conv blocks) and a fully-connected output layer in the end. Each ST-Conv block contains two temporal gated convolution layers and one spatial graph convolution layer in the middle. The residual connection and bottleneck strategy are applied inside each block. The input $v_{t-M+1}, ..., v_{t}$ is uniformly processed by ST-Conv blocks to explore spatial and temporal dependencies coherently. Comprehensive features are integrated by an output layer to generate the final prediction $\hat{v}$.
• Figure 3: PeMS sensor network in District 7 of California (left), each dot denotes a sensor station; Heat map of weighted adjacency matrix in PeMSD7(M) (right).
• Figure 4: Speed prediction in the morning peak and evening rush hours of the dataset PeMSD7.
• Figure 5: Test RMSE versus the training time (left); Test MAE versus the number of training epochs (right). (PeMSD7(M))