A Big Data Demand Estimation Framework for Modelling Urban Congested Networks

Abstract

This paper deals with the problem of estimating daily mobility flows using different sources of data, and in particular from mobile devices such as mobile phones and floating car data. First, the dynamic origin-destination (OD) estimation problem is presented in its classical bi-level formulation, and opportunities and challenges of introducing new data types with respect to the classical fixed sensors are discussed. We show how mobile phone data can be used to better estimate the structure of the demand matrix, both temporally (i.e. the daily generated flows from each zone) and spatially (i.e. distributing the flows on the different OD pairs). Then, floating car data and traffic counts can be used to further distribute the demand on the available modes and routes, according to the classical 4-step procedure. During this phase, a behavioural modelling approach is used, according to traditional dynamic user equilibrium using a joint route and departure time choice model. Floating car data information is used to estimate speed profiles at all links where information is available, and for route travel times, which feed the utility-based models. A two-step approach is then proposed to solve the problem for large scale networks, in which the total demand is first generated, and then equilibrium is calculated through a dynamic traffic assignment model. The effectiveness and reliability of the proposed modelling framework is then shown on a realistic case study involving the multimodal network of Luxembourg City and its surrounding, and is compared to the traditional bi-level formulation solved using the Generalised Least Square Estimation. The comparison shows how the two-step approach is more robust in generating realistic daily OD flows, and in exploiting the information collected from mobile sensors.