Abstract:Forecasting traffic flows is a central task in intelligent transportation system management. Graph structures have shown promise as a modeling framework, with recent advances in spatio-temporal modeling via graph convolution neural networks, improving the performance or extending the prediction horizon on traffic flows. However, a key shortcoming of state-of-the-art methods is their inability to take into account information of various modalities, for instance the impact of maintenance downtime on traffic flows. This is the issue we address in this paper. Specifically, we propose a novel model to predict traffic speed under the impact of construction work. The model is based on the powerful attention-based spatio-temporal graph convolution architecture but utilizes various channels to integrate different sources of information, explicitly builds spatio-temporal dependencies among traffic states, captures the relationships between heterogeneous roadway networks, and then predicts changes in traffic flow resulting from maintenance downtime events. The model is evaluated on two benchmark datasets and a novel dataset we have collected over the bustling Tyson's corner region in Northern Virginia. Extensive comparative experiments and ablation studies show that the proposed model can capture complex and nonlinear spatio-temporal relationships across a transportation corridor, outperforming baseline models.
What problem does this paper attempt to address?
### What problem does this paper attempt to solve?
This paper aims to solve a key problem in traffic flow prediction, that is, existing methods cannot fully consider information of different modalities, especially the impact of road maintenance downtime on traffic flow. Specifically, the current state - of - the - art traffic flow prediction models have limitations when dealing with complex and multi - source information and cannot accurately predict traffic speed changes during construction.
#### Main problems:
1. **Multi - modal information fusion**: Most of the existing traffic flow prediction models can only handle a single type of data (such as traffic speed) and cannot comprehensively consider the impact of multiple factors (such as construction, weather, etc.) on traffic flow.
2. **Impact of construction downtime**: Construction downtime will not only lead to traffic congestion but also bring significant safety hazards. However, most existing models cannot effectively incorporate these factors into the prediction model, resulting in a decline in prediction performance.
3. **Capture of spatio - temporal correlations**: Traffic flow has complex spatio - temporal correlations. Especially in large - scale traffic networks, how to effectively capture these correlations is a challenge.
#### Solutions:
To solve the above problems, the author proposes a Multi - Channel Attention - Based Spatio - Temporal Graph Convolutional Networks (GCN - RWZ). The main features of this model include:
- **Multi - channel information fusion**: By introducing multi - channel feature maps, the model can simultaneously process multiple data sources such as traffic speed and construction information, and integrate this information together through the feature fusion function.
- **Spatio - temporal attention mechanism**: The model uses a multi - head attention mechanism, which can better capture the complex spatio - temporal correlations in the traffic network.
- **Modeling of construction impact**: In particular, the model introduces a feature map of the construction work zone to explicitly model the impact of construction on traffic flow.
#### Experimental verification:
The author conducted experiments on three real - world datasets, including a newly collected dataset in Tyson's Corner area. The experimental results show that the GCN - RWZ model can effectively capture complex non - linear spatio - temporal relationships, and in the case of including construction information, the prediction performance is significantly better than other benchmark models.
Through these improvements, the GCN - RWZ model can provide more accurate flow predictions in more complex and dynamic traffic environments, thereby helping traffic management departments optimize construction arrangements, reduce traffic congestion and improve traffic safety.
---
### Formula summary
- **Definition of construction feature map**:
\[
X_c^v=\max\left(0, 1-\left(\frac{\text{dis}(v_i, v_j)}{\lambda}\right)^2\right)
\]
where $\text{dis}(v_i, v_j)$ represents the geometric distance between nodes $v_i$ and $v_j$, and $\lambda$ is a hyperparameter used to adjust the weight of construction impact.
- **Feature fusion function**:
\[
\hat{X}_s = W_s\odot X_s+W_c\odot X_c
\]
where $W_s$ and $W_c$ are learning parameters, representing the impact degree of construction downtime on the predicted traffic state respectively.
- **Graph convolution operation**:
\[
(x * g)_G = U g_\theta U^T x
\]
where $U$ is the eigenvector matrix of the Laplacian matrix, and $g_\theta$ is a learnable convolution kernel.
- **Chebyshev polynomial expansion**:
\[
g_\theta(\Lambda)\approx\sum_{k = 0}^K\theta_k T_k(\hat{\Lambda})
\]
where $T_k$ is the Chebyshev polynomial.