Prediction of Crowd Flow in Key Areas Based on a Spatial-temporal Convolutional Model

During the epidemic, the flow and aggregation of the population have objectively increased the risk of epidemic transmission and the difficulty of prevention and control. To further grasp the movement and aggregation of people and do a good job in the prevention and control of emergency epidemics, this paper proposes a model based on spatial-temporal convolutional networks to predict the population density in key areas. The model is mainly composed of Graph Convolutional Network (GCN) and Gate Recurrent Unit (GRU). Compared with general time series prediction problems, crowd density prediction has temporal and spatial dependencies. Traditional time series modeling ideas cannot deal with these characteristics effectively. The abstraction is a graph structure, which fully exploits the spatial dependence of crowd flow. Besides, this work uses the GRU model to extract the temporal correlation of crowd flow for accurately predicting future crowd density.