Hierarchical graph construction for station-free traffic demand prediction

Graph neural networks (GNNs) have been widely applied in traffic demand prediction, and traffic modes can be divided into station-based modes, such as buses and subways, and station-free modes, such as ride-hailing and bike-sharing. Existing research in traffic graph construction primarily relies on the distribution of actual traffic stations to build graphs based on the road network. However, station-free traffic demand prediction cannot utilize real traffic stations, and the complexity and heterogeneous distribution of data further introduce significant challenges. As a result, effectively leveraging this information for constructing graph structures has become a major challenge. To tackle these, this paper introduces a novel hierarchical graph construction method tailored to station-free traffic mode. We propose a novel density-based clustering algorithm (HDPC-L) to determine the flexible positioning of nodes in the graph, overcoming the computational bottlenecks of traditional clustering algorithms and enabling effective handling of large-scale datasets. Furthermore, we extract valuable information from ridership data to initialize the edge weights of GNNs. Comprehensive experiments on two real-world datasets, the Shenzhen bike-sharing dataset and the Haikou ride-hailing dataset, show that the method significantly improves the performance of the model. On average, our models show an improvement in accuracy of around 17.74% and 13.41% on the two datasets. Additionally, it significantly enhances computational efficiency, reducing training time by approximately 7.64% and 27.16% on the two datasets. We make our code available at https://github.com/houjinyan/HDPC-L-ODInit.