Accurate prediction of short-term OD Matrix (i.e. the distribution of passenger flows from various origins to destinations) is a crucial task in metro systems. It is highly challenging due to the constantly changing nature of spatiotemporal factors and the recent data collection problem. Recently, some deep learning-based models have been proposed for OD Matrix forecasting in ride-hailing or high way traffic scenarios. However, these models can not sufficiently capture the complex spatiotemporal correlations among stations in metro networks due to different prior knowledge and contextual settings. In this paper, we propose a model, namely Multi-Scale STATCN, to address OD metro matrix prediction. Specifically, it first proposes a data-driven method to solve the recent data collection problem. Then, it captures the dynamic spatial dependency in OD flows among different stations by a global self-attention mechanism. Three temporal convolutional networks are leveraged to capture three temporal trends in OD flow, i.e. recent trend, daily trend, weekly trend. Extensive experiments on three large-scale metro datasets demonstrate the superiority of our model over other competitors.
View on arXiv