Fig. 5
From: Spatio-temporal transformer and graph convolutional networks based traffic flow prediction
Schematic of the model. This figure shows the TDMGCN model diagram, which is mainly composed of spatiotemporal embedding layers and L identical encoders and decoders. The function of the spatiotemporal embedding layer is to enable the model to learn spatiotemporal information. The encoder and decoder are composed of a local trend-aware self-attention mechanism and an adaptive graph convolution based on multiple graphs. The model generates prediction results through autoregression, using the data generated in the previous steps as additional input when predicting the next value.
