Real-time Predictive Analytics for Improving Public Transportation Systems' Resilience

TL;DR

This paper presents a hybrid real-time predictive analytics system that enhances public transportation resilience by proactively managing disruptions and improving passenger trip planning using sensor data and predictive models.

Contribution

It introduces a novel hybrid data/model-driven decision support system utilizing real-time predictive models for transit management and passenger assistance.

Findings

Improved prediction accuracy for station arrivals and trip origins.

Enhanced system responsiveness to disruptions.

Better passenger trip decision support.

Abstract

Public transit systems are a critical component of major metropolitan areas. However, in the face of increasing demand, most of these systems are operating close to capacity. Under normal operating conditions, station crowding and boarding denial are becoming a major concern for transit agencies. As such, any disruption in service will have even more severe consequences, affecting huge number of passengers. Considering the aging infrastructure of many large cities, such as New York and London, these disruptions are to be expected, amplifying the need for better demand management and strategies to deal with congested transit facilities. Opportunistic sensors such as smart cards (AFC), automatic vehicle location systems (AVL), GPS, etc. provide a wealth of information about system's performance and passengers' trip making patterns. We develop a hybrid data/model-driven decision support system, using real-time predictive models, to help transit operators manage and respond proactively to disruptions and mitigate consequences in a timely fashion. These models include station arrival and origin-destination predictions in real-time to help transit agencies, and predictive information systems for assisting passengers' trip making decisions.