Songhua Hu

Songhua Hu

CrowdSensing, Human Mobility, GeoAI, Network Science

Large-scale spatiotemporal network forecasting

Spatiotemporal mobility forecasting is challenging due to complex spatiotemporal dependencies and heterogeneous external effects. My ongoing research focuses on large-scale spatiotemporal network forecasting using deep learning methods. I proposed a Multi-graph Multi-head Adaptive Temporal Graph Convolutional Network (Multi-ATGCN), a general deep learning framework for citywide multi-step human mobility forecasting. Experiments on two real-world tasks demonstrate its steady performance improvement over state-of-the-art baselines. I also propose a hierarchical activity-based framework for simultaneously predicting the activity, time, and location of the next trip for each device.

Journals

  1. Hu, Songhua, Chenfeng Xiong. High-dimensional population flow time series forecasting via an interpretable hierarchical transformer, Transportation Research Part C: Emerging Technologies 146 (2023): 103962.

Working Papers

  1. Hu, Songhua, Yiqun Xie, Chenfeng Xiong, Paul Schonfeld, Multi-ATGCN: A multi-graph multi-head adaptive temporal graph convolutional network for multivariable crowd inflow forecasting (2023). (Intended for KDD 2023)
  2. Hu, Songhua, Yiqun Xie, Peng Chen, Paul Schonfeld, Nationwide spatiotemporal population flow forecasting via temporal fusion graph convolutional neural network: a comparative analysis. (2023). (Intended for IEEE Transactions on Intelligent Transportation Systems)