An empirical study on the stochastic long-term travel demands of a large-scale metro network

Abstract

The widely-existed uncertainty of origin–destination (OD) demand in transportation networks has attracted extensive attention. Most characterizations or modeling of stochastic OD demands in networks assume a homogenous probability distribution, but it lacks empirical studies of large-scale networks to justify this assumption. Given that the long-term continuous automatic fare collection data of metro networks can provide complete OD passenger demand information, this study takes the Shanghai Metro network as an example to empirically examine the stochasticity characteristics of OD passenger demands of metro network. Based on the morning peak OD demand data for 250 weekdays, a local outlier factor method is used to identify and remove outliers in the data. A clustering method is used to cluster the OD pairs, study the fluctuation and distribution characteristics of the OD passenger demands, and select the optimal distribution type through goodness-of-fit indices. The results show that 1) the coefficients of variation of morning peak OD demands in the network are mainly distributed in the range of 0.2–0.6, and different OD pairs have different fluctuations, and the degree of demand fluctuation decreases as the mean increases; 2) the probability distribution types of OD demands based on statistical characteristics are heterogeneous; and 3) the optimal distribution type of OD demands is Poisson, Lognormal/Gamma, and Normal for OD pairs characterized by a small mean and right-skewness, a small mean and skewness close to 0, and a large mean, respectively. In contrast to the simple average-based data processing of OD passenger demand in metro networks, this paper presents a new perspective of mining long-term continuous data to understand the inherent stochasticity of OD passenger demands. The results can provide more realistic and practical inputs and assumptions for theoretical research on stochastic OD demands in metro networks.