NATIONWIDE PASSENGER FLOW FORECASTING: A SPATIOTEMPORAL DEEP LEARNING APPROACH

Forecasting nationwide city-level passenger flow is vital for passenger flow management and effective allocation of national transportation resources. However, it is a challenging task in practice as nationwide city-level passenger flow is affected by complex factors including the irregular shapes of cities, intra-city correlations (inflow relates to outflow), spatial and temporal dependencies, and exogenous factors. To cope with aforementioned challenges, this paper proposes a deep learning approach named Multi-Fused Residual Networks (MF-ResNet) to simultaneously forecast inflow and outflow of city-level passengers in China. The proposed method is validated by the real-world travel demand data provided by Tencent Location Big Data Platform, and the experimental results show that MF-ResNet can well capture the spatiotemporal dependencies. Through comparison, MF-ResNet outperforms state-of-art baselines in terms of forecasting accuracy.