Bus network plays a critical role in urban transportation affecting the use of private vehicles, traffic congestion, and urban accessibility. The accurate prediction of bus passenger flow is key to improving transit passenger experience and increasing the efficiency of bus network operations. In line with recent advances in deep learning for passenger flow prediction, graph neural networks (GNNs) have become increasingly popular due to their ability to account for the network structure between stops. Existing GNN-based models for bus passenger flow prediction, however, face several limitations. First, they do not take into account some distinctive characteristics of bus networks, such as their coexistence with vehicular traffic and their high sensitivity to urban traffic conditions. Moreover, sequence prediction models that have been widely applied to multistep passenger flow prediction suffer from a critical issue, called “exposure bias.” This results in the propagation and accumulation of errors through prediction steps while making predictions for farther time horizons. To address these issues, this study presents the Traffic-Aware multistep Graph Neural Network (TMS-GNN) model with Scheduled Sampling, a graph-based deep-learning framework designed to forecast multistep bus passenger flows at individual stops across a bus network. The model takes into account factors such as bus stop connectivity, urban traffic impacts, and multi-dimensional temporal patterns; and addresses exposure bias by employing a curriculum learning strategy called Scheduled Sampling. The comparison between the proposed model and other popular baseline models on two real-world networks with different geographical and urban patterns in Canada and USA shows that TMS-GNN outperforms the baselines in both the overall network-wide task, as well as multistep prediction. Also, to verify the contribution of the proposed components of the model, an ablation study is conducted. The results of the ablation study validate the design choices as well.

Graph neural networks (GNNs) have been extensively used in a wide variety of domains in recent years. Owing to their power in analyzing graph-structured data, they have become broadly popular in intelligent transportation systems (ITS) applications as well. Despite their widespread applications in different transportation domains, there is no comprehensive review of recent advancements and future research directions that covers all transportation areas. Accordingly, in this survey, for the first time, we provide an overview of GNN studies in the general domain of ITS. Unlike previous surveys, which have been limited to traffic forecasting problems, we explore how GNN frameworks have evolved for different ITS applications, including traffic forecasting, demand prediction, autonomous vehicles, intersection management, parking management, urban planning, and transportation safety. Also, we micro-categorize the studies based on their transportation application to identify domain-specific research directions, opportunities, and challenges, which have been missing in previous surveys. Moreover, we identify unique and undiscussed research opportunities and directions, which is the result of reviewing a wide range of transportation applications. The neglected role of edge and graph learning in ITS applications, developing multi-modal models, and exploiting the power of unsupervised and reinforcement learning methods for developing more powerful GNNs are some examples of such new discussions in this survey. Finally, we have identified popular baseline models and datasets in each transportation domain, which facilitate the development and evaluation of future GNN-based frameworks.

Predicting short-term passenger flows in bus networks is crucial to improving the overall performance of such systems and increasing their attractiveness. This study develops a graph neural network-based framework for multi-step passenger flow prediction specifically designed for bus networks to capture their unique characteristics. We propose the Multi-step Multi-component Graph Convolutional Long Short-Term Memory (Multi-GCN-LSTM) model, which uses 1) a proximity matrix in addition to an adjacency matrix to consider the effects of vehicular traffic and link-level distances; 2) Scheduled Sampling for multi-step prediction, which prevents error propagation across prediction steps; and 3) a novel fusion mechanism for considering time-varying spatial and temporal correlations among passenger flow data based on recent, daily, and weekly travel patterns. This model is validated using real-world data collected from the Laval bus network. Also, benchmarking the established model against state-of-the-art baselines indicated its competency.