A Variational Graph Convolution Network with Normalizing Flows for Passenger Flow Prediction

Passenger flow prediction plays a crucial role in intelligent transportation systems, aiming to facilitate affordable and eco-friendly public transportation. In this context, we introduce a comprehensive framework known as the Variational Graph Convolution Network with Normalizing Flows for Passenger Flow Prediction (VGConvNF). By combining variational graph convolution with the acquisition of a probabilistic latent variable that encodes changes in passenger flows, this innovative approach enables the modeling of dynamic passenger flows and bus stop graph structures. Furthermore, we use the power of normalizing flows to extend the range of possible distributions in order to enhance the posterior approximation. A real-world dataset of passenger flows was used to evaluate the effectiveness of our proposed method. Results demonstrate the effectiveness of our approach in extending graph neural networks with probabilistic node embeddings and enriching the core graph neural networks. This framework significantly improves the accuracy and reliability of passenger flow prediction, thus contributing to the development of efficient and sustainable transportation systems.

I. Introduction

Transportation problems in big cities have became more and more severe with rapid urbanization. A low-cost, en-vironmentally friendly method of public transportation is a necessity for residents of densely populated cities. A city bus transportation system is a vital component of a public transportation network. Weather, traffic congestion, or public events can cause sudden changes in passenger flow, resulting in more passengers waiting for buses than usual. Thus, short-term, highly accurate passenger flow prediction is crucial for effective system management. In passenger flow prediction, the challenge is to capture both spatial and temporal depen-dencies of passenger flows at all bus stops at the same time. This challenge is heightened by the complex structure of road crossings and lanes, highly nonlinear correlations between passenger flow at different bus stops, and the non-deterministic nature of passenger flow.