A graph spatial-temporal model for predicting population density of key areas

Predicting the population density of key areas of the city is crucial. It helps reduce the spread risk of Covid-19 and predict individuals' travel needs. Although current researches focus on using the method of clustering to predict the population density, there is almost no discussion about using spatial-temporal models to predict the population density of key areas in a city without using actual regional images. We abstract 997 key areas and their regional connections into a graph structure and propose a model called Word Embedded Spatial-temporal Graph Convolutional Network (WE-STGCN). WE-STGCN is mainly composed of the Spatial Convolution Layer, the Temporal Convolution Layer, and the Feature Component. Based on the data set provided by the DataFountain platform, we evaluate the model and compare it with some typical models. Experimental results show that WE-STGCN has 53.97% improved to baselines on average and can commendably predicting the population density of key areas.

Automated summary

Predicting the population density of key areas in a city is crucial for reducing the spread risk of Covid-19 and predicting travel needs. The proposed WE-STGCN model significantly improves the accuracy of population density prediction, outperforming typical models by 53.97% on average.