Abstract:Traffic forecasting is a particularly challenging application of spatiotemporal forecasting, due to the time-varying traffic patterns and the complicated spatial dependencies on road networks. To address this challenge, we learn the traffic network as a graph and propose a novel deep learning framework, Traffic Graph Convolutional Long Short-Term Memory Neural Network (TGC-LSTM), to learn the interactions between roadways in the traffic network and forecast the network-wide traffic state. We define the traffic graph convolution based on the physical network topology. The relationship between the proposed traffic graph convolution and the spectral graph convolution is also discussed. An L1-norm on graph convolution weights and an L2-norm on graph convolution features are added to the model's loss function to enhance the interpretability of the proposed model. Experimental results show that the proposed model outperforms baseline methods on two real-world traffic state datasets. The visualization of the graph convolution weights indicates that the proposed framework can recognize the most influential road segments in real-world traffic networks.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to conduct accurate traffic state prediction in traffic networks. Specifically, due to the fact that traffic patterns change over time and the spatial dependencies in road networks are complex, traffic prediction has become a particularly challenging application in spatio - temporal prediction. To meet this challenge, the authors propose a new deep - learning framework - Traffic Graph Convolutional Long Short - Term Memory Neural Network (TGC - LSTM), aiming to learn the interactions in road networks and predict the traffic states of the entire network. ### Main Problems 1. **Time - Varying Traffic Patterns**: Parameters such as traffic flow and speed change over time, and this temporal instability increases the difficulty of prediction. 2. **Complex Spatial Dependencies**: The traffic state in road networks is affected not only by its own historical data but also by the traffic states of adjacent road sections. This complex spatial dependency makes it difficult for traditional statistical methods to handle. ### Solutions The authors propose the following solutions: 1. **Traffic Graph Convolution Operator**: Define traffic graph convolution based on the physical network topology to capture local features in traffic networks. 2. **TGC - LSTM Model**: Combine graph convolution and Long Short - Term Memory (LSTM) network to learn the complex spatio - temporal dependencies in traffic data. 3. **Regularization Term**: Add L1 - norm and L2 - norm to the loss function of the model to enhance the interpretability of the model. ### Main Contributions of the Model 1. **Traffic Graph Convolution Operator**: Propose a traffic graph convolution operator that adapts to the physical characteristics of traffic networks and can extract comprehensive features. 2. **TGC - LSTM Model**: Propose a traffic graph convolution LSTM neural network that can learn the complex spatio - temporal dependencies in traffic data. 3. **Regularization Term**: By adding L1 - norm and L2 - norm to the loss function, make the learned local graph convolution features more consistent and interpretable. 4. **Public Data Set**: Provide real - world traffic speed data for research, including the graph structure of traffic networks, which promotes further research. ### Mathematical Formulas - **Adjacency Matrix and Neighborhood Matrix**: \[ \tilde{A}=A + I \] \[ \tilde{A}_k=\min((A + I)^k, 1) \] - **Free - Flow Reachability Matrix**: \[ \mathcal{F}\mathcal{F}\mathcal{R}_{i,j}= \begin{cases} 1 & \text{if } S_{i,j}^{\mathcal{F}\mathcal{F}}\Delta t - D_{i,j}\geq0 \\ 0 & \text{otherwise} \end{cases} \] - **Traffic Graph Convolution Operation**: \[ G\mathcal{C}_t^k=(W_{gc}^k\odot\tilde{A}_k\odot\mathcal{F}\mathcal{F}\mathcal{R})x_t \] - **Multi - Order Traffic Graph Convolution Features**: \[ \mathbf{G}\mathcal{C}_t^{(K)}=[G\mathcal{C}_t^1, G\mathcal{C}_t^2,\ldots, G\mathcal{C}_t^K] \] - **LSTM Gating Mechanism**: \[ f_t=\sigma_g(W_f\cdot\mathbf{G}\mathcal{C}_t^{(K)}+U_f\cdot h_{t - 1}+b_f) \] \[ i_t=\sigma_g(W_i\cdot\mathbf{G}\mathcal{C}_t^{(K)})

Traffic Graph Convolutional Recurrent Neural Network: A Deep Learning Framework for Network-Scale Traffic Learning and Forecasting

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

Graph learning-based spatial-temporal graph convolutional neural networks for traffic forecasting

3D Graph Convolutional Networks with Temporal Graphs: A Spatial Information Free Framework For Traffic Forecasting

Constructing Geographic and Long-term Temporal Graph for Traffic Forecasting

Attention Based Graph Covolution Networks for Intelligent Traffic Flow Analysis

A deep learning traffic flow prediction framework based on multi-channel graph convolution

Hierarchical Spatio–Temporal Graph Convolutional Networks and Transformer Network for Traffic Flow Forecasting

T-GCN: A Temporal Graph ConvolutionalNetwork for Traffic Prediction

Spatial dynamic graph convolutional network for traffic flow forecasting

T-GCN: A Temporal Graph Convolutional Network for Traffic Prediction

TSSRGCN: Temporal Spectral Spatial Retrieval Graph Convolutional Network for Traffic Flow Forecasting.

Optimized Graph Convolution Recurrent Neural Network for Traffic Prediction

Graph attention temporal convolutional network for traffic speed forecasting on road networks

Spatial-temporal gated graph convolutional network: a new deep learning framework for long-term traffic speed forecasting

Local-Global Spatial-Temporal Graph Convolutional Network for Traffic Flow Forecasting

Attention-based Spatial-Temporal Graph Convolutional Recurrent Networks for Traffic Forecasting

Probabilistic spatio-temporal graph convolutional network for traffic forecasting

Spatio-Temporal Heterogeneous Graph-Based Convolutional Networks for Traffic Flow Forecasting

Spatial-Temporal Residual Multi-Graph Convolution Network for Traffic Forecasting