An improved convolutional network capturing spatial heterogeneity and correlation for crowd flow prediction

Crowd flow prediction plays an important role in urban management and public safety. However, the existing prediction models still have some shortcomings in capturing spatial heterogeneity and multi-scale spatial correlation. To fulfill the research gaps, this paper proposes an improved convolutional network (SHC-Net). The proposed SHC-Net model improves the existing models by capturing the spatial heterogeneity of the temporal patterns of crowd flow, considering both global and local spatial correlations simultaneously, and combining external factors and spatiotemporal features to consider the heterogeneous impact of external factors on crowd flows. We conduct experiments on two real large-scale datasets, and the results show that our model consistently outperforms the state-of-the-art baselines.