Post a summary post: Paper reading record
Paper link: "T-GCN: A Temporal Graph Convolutional Network for Traffic Prediction"
1. Summary
Accurate and real-time traffic forecasts play an important role in intelligent transportation systems, and are of great significance to urban traffic planning, traffic management and traffic control. However, due to the constraints of the urban road network topology and dynamic changes, traffic prediction has always been considered an "open" scientific problem. In order to capture the spatial and temporal dependence at the same time, we propose a new neural network-based traffic prediction method-Time Graph Convolutional Network (T-GCN) model, which combines graph convolutional network (GCN) and gating The regression unit (GRU) is combined. Specifically, GCN is used to learn complex topological structures to capture spatial dependence, and a gated regression unit is used to learn dynamic changes of traffic data to capture temporal dependence. Then, the T-GCN model is applied to traffic prediction based on the urban road network. Experiments show that our T-GCN model can obtain temporal and spatial correlations from traffic data, and the prediction results are better than the world's most advanced traffic data set baseline.
The main content of the article
A traffic prediction method based on neural network is proposed—time graph convolutional network (T-GCN) model, which combines graph convolutional network (GCN) and gated recurrent unit (GRU). Applied to traffic prediction based on urban road network, experiments show that the prediction result of T-GCN model is better than the benchmark model.
2. Conclusion
- A new neural network-based traffic prediction method is developed, called T-GCN, which combines GCN and GRU.
- A graph network is used to model the urban road network. The nodes on the graph represent the roads, and the edges represent the connection relationships between the roads. The traffic information on the roads is described as the attributes of the nodes on the graph. On the one hand, GCN is used to capture the spatial topological structure of the graph to obtain the spatial dependence of the graph; on the other hand, the GRU model is introduced to capture the dynamic changes of node attributes and obtain the time dependence of node attributes. Finally, the T-GCN model is used to deal with spatio-temporal traffic prediction tasks.
- It is evaluated on two real traffic data sets and compared with HA model, ARIMA model, SVR model, GCN model and GRU model. The T-GCN model performs better in different prediction intervals.
- The T-GCN model successfully captures spatio-temporal features from traffic data and can be applied to other spatio-temporal tasks.
Three, T-GCN
Time graph convolutional network.
- The T-GCN model integrates graph convolutional networks and gated recursive units. The graph convolutional network is used to capture the road network topology and model the spatial dependence of the road network. The gated recursive unit is used to capture the dynamic changes of road traffic data and model the time dependence. The T-GCN model can also be applied to other spatiotemporal prediction tasks.
- ) The prediction results of the T-GCN model show that under different prediction levels, the T-GCN model is in a stable state, which shows that the T-GCN model can not only achieve short-term predictions, but also can be used for long-term traffic prediction tasks (15 minutes 30 Minutes 45 minutes 60 minutes).
- Two real traffic data sets are used to evaluate the method. The results show that compared with all baseline methods, the prediction error of this method is reduced by about 1.5%-57.8%, which proves the superiority of the T-GCN model in traffic prediction.
Regardless of the prediction time range, the T-GCN model can always achieve better prediction results. The T-GCN model can capture the temporal and spatial characteristics of road traffic information and get the trend of changes in road traffic information. In addition, the T-GCN model detects the start and end time of the traffic peak period, and makes predictions similar to the actual traffic speed. These properties help predict traffic congestion and other traffic phenomena.