A Bayesian Markov Model for Station-Level Origin-Destination Matrix Reconstruction

Abstract

This paper tackles Origin-Destination (OD) matrix reconstruction at a station level, which consists in computing the volume of passengers traveling between two different stations on a public transportation network. This information is critical for the transport operator to compute various indicators concerning the network’s state and performance such as vehicle occupancy and travelers’ behavior. Trip reconstruction for smart card holders, whose history of validations is available, has been thoroughly investigated in prior work. Conversely, trip reconstruction for non smart card holders has received less attention, mainly due to the difficulty of obtaining ground truth data. Among recent work in this domain, very few contributions have tackled large networks in their entirety, with millions of validations over a month and the computational challenges that come with it. In this work, we present a new Bayesian Markov Model for OD matrix reconstruction. The novelty of our model lies in its scalability and the fact that it uses all available data, including Automated Fare Collection ( i.e. smart card holders) data and Automatic Passenger Counting data ( i.e. data from counting sensors), to accurately infer the trips’ distribution. Moreover, the proposed approach produces proper OD matrices while taking into account sensor noise and fraud. We empirically establish the relevance, robustness, and accuracy of the proposed method compared to the popular trip chaining algorithm and a previous Markov based approach on real-world, large-scale industrial datasets for two transportation networks in major cities.