Crowd Flow Prediction for Irregular Regions with Semantic Graph Attention Network

It is essential to predict crowd flow precisely in a city, which practically partitioned into irregular regions based on road networks and functionality. However, prior works mainly focus grid-based prediction, where city divided many regular grids. Although Convolutional Neural Netwok (CNN) powerful capture spatial dependence from Euclidean data, it fails tackle non-Euclidean reflect the correlations among regions. Besides, fail jointly hierarchical spatio-temporal both Finally, are time-varying functionality-related. combination of dynamic semantic attributes ignored by related works. To address above challenges, this article, we propose novel model prediction task for First, employ CNN Graph Network (GNN) micro macro regions, respectively. Further, think highly inter-region location-aware time-aware graph attention mechanism named Semantic Attention (Semantic-GAT), node attribute embedding multi-view reconstruction. Extensive experimental results two real-life datasets demonstrate that our outperforms 10 baselines reducing error around 8%.