Online Car-Hailing Origin-Destination Forecast Based on a Temporal Graph Convolutional Network

Abstract

Online car-hailing has become an indispensable transportation means for residents. The short-term origin and destination (OD) prediction of online car-hailing trips is conducive to understanding the inflow and outflow of online car-hailing trips in a region and provides data support for the delivery and scheduling of vehicles. Accordingly, this article takes the data of car-hailing trips in the central area of Haikou, China, as the research data; makes an in-depth analysis of the regularities of car-hailing trips; and divides the central area of Haikou into 84 grids with a length of 3 km. This article constructs three adjacency matrices, Am01, Sam, and Amn, to reflect the complex spatial relationships of the OD matrixes of online car-hailing from different perspectives. Then, a model, based on the graph convolutional network (GCN) and gated recurrent unit, denoted as the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">temporal GCN</i> ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">T-GCN</i> ), is introduced for the grid-based short-term OD prediction. The case study in Haikou shows that T-GCNs based on the three adjacency matrices are better than other models, wherein the Amn-based T-GCN is more consistent with the OD flows’ spatial relationship, achieves the best prediction performance, and shows that there exists a proportional relationship between flows on different OD pairs. The application of the research results is beneficial for the car-hailing platform to perform the dynamic scheduling of vehicles in advance, further improving the operating efficiency and reducing the waiting time of passengers so as to effectively alleviate the problem of the imbalance between the supply and demand of online car-hailing travel.