An adaptive spatio-temporal graph neural network for traffic prediction

For the problem that the neural network can′t capture the long-term traffic information in the spatial dimension, the new neural network structure proposed in the past can′t capture the complex traffic data in the spatial dimension. Through adaptive graph convolutional network, the specific state of nodes is automatically captured and the interdependence between different nodes is automatically inferred to extract the complete spatial features of traffic data. Then, the time characteristics of traffic data are captured by the time memory module in the spatio-temporal short-term memory network, and the short, medium and long-term time dependence is simulated.