Predicting origin-destination ride-sourcing demand with a\n spatio-temporal encoder-decoder residual multi-graph convolutional network

Abstract

With the rapid development of mobile-internet technologies, on-demand\nride-sourcing services have become increasingly popular and largely reshaped\nthe way people travel. Demand prediction is one of the most fundamental\ncomponents in supply-demand management systems of ride-sourcing platforms. With\naccurate short-term prediction for origin-destination (OD) demand, the\nplatforms make precise and timely decisions on real-time matching, idle vehicle\nreallocations and ride-sharing vehicle routing, etc. Compared to zone-based\ndemand prediction that has been examined by many previous studies, OD-based\ndemand prediction is more challenging. This is mainly due to the complicated\nspatial and temporal dependencies among demand of different OD pairs. To\novercome this challenge, we propose the Spatio-Temporal Encoder-Decoder\nResidual Multi-Graph Convolutional network (ST-ED-RMGC), a novel deep learning\nmodel for predicting ride-sourcing demand of various OD pairs. Firstly, the\nmodel constructs OD graphs, which utilize adjacent matrices to characterize the\nnon-Euclidean pair-wise geographical and semantic correlations among different\nOD pairs. Secondly, based on the constructed graphs, a residual multi-graph\nconvolutional (RMGC) network is designed to encode the contextual-aware spatial\ndependencies, and a long-short term memory (LSTM) network is used to encode the\ntemporal dependencies, into a dense vector space. Finally, we reuse the RMGC\nnetworks to decode the compressed vector back to OD graphs and predict the\nfuture OD demand. Through extensive experiments on the for-hire-vehicles\ndatasets in Manhattan, New York City, we show that our proposed deep learning\nframework outperforms the state-of-arts by a significant margin.\n