Finding Most Reliable Path With Extended Shifted Lognormal Distribution

Abstract

Travel time uncertainty may cause late arrival and impose a high penalty on travelers. There is a growing interest in modeling travel time uncertainty to optimize the reliability of travel time at the path and network level. Real data analysis finds that the influence factors, including day-of-week, holidays, time-of-day, road grades, traffic states, and so on, often reduce the cumulative probability of travel time even in the same facility type (the same lane number and the same divided type). Thus, a novel aggregate approach is proposed to classify the travel time data based on these influence factors. The distribution with the new aggregate approach is defined as the extended shifted lognormal (ESLN) distribution. KS test indicates that the ESLN distribution can effectively describe travel time, and outperforms normal, lognormal, gamma, and beta distribution. Travel time correlations are calculated between new aggregate groups, which can effectively reduce the complexity compared with the link to link correlations. Finally, the ESLN distribution is used to find the most reliable path in a real-world large-scale network. The comparison results between ESLN distribution and shifted lognormal (SLN) distribution show the effectiveness and improvement of the proposed method in finding the most reliable path.