Correlation analysis of day-to-day origin-destination flows and traffic volumes in urban networks

Abstract

The trip patterns on an urban network can be represented by two main variables: origin-destination flows (OD flows), defined as the number of trips between two locations over a given time period, and traffic volumes, defined as the number of vehicles that cross a street over a given time interval. Past research on the dynamic of traffic assignment and OD estimation suggested that the traveler's decisions vary on a day-to-day basis and that their most recent decisions may affect their current travel decisions. Based on these assumptions, this study analyzed the autocorrelation of a set of day-to-day series of traffic volumes and OD flows generated from a large collection of traffic sensors, identifying the data's correlation structure over different locations and OD pairs in an urban network. To this end, a method for data treatment of the 2017 dataset from the traffic monitoring system of Fortaleza, Brazil, was employed, which consisted in the following major steps: data cleaning due to equipment failure, definition of traffic profiles for typical and atypical months, definition of daily traffic periods, selection of suitable devices to obtain OD flows, and detection of outliers in the time series. The traffic profiles and the daily traffic periods were defined by applying clustering techniques. The analysis of autocorrelation was performed after controlling for seasonal effects in the data by applying regression analysis. This study contributes to understand how the dynamic of trip patterns varies over space due to the spatial distribution of the city's activities and the network's spatial centrality. The analysis of 144 sets of traffic volumes throughout 2017 suggests that the autocorrelation of traffic volumes should be higher in congested central areas where multiple options of route are available. It seems that, for large congested networks, which present many uncertain factors (e.g., accidents, variable weather, multiple paths, etc.), part of the users do not have complete knowledge of the network's performance, and must rely on experience and habit to decide their routes, especially at more centralized locations of the network. The analysis of serial correlation in the series of sample OD flows between regions showed that the city's central area, where more commercial and service-related activities take place, seems to influence the dynamic of OD flows, probably due to the occurrence of more non-commuting trips to the central area of the city.