Passenger itinerary inference model for congested urban rail networks

Abstract

The paper develops a data-driven approach for the inference of passenger itineraries in urban heavy rail systems, the Passenger Itinerary Inference Model (PIIM). In light of increasing demand and crowding, the model can be used to assess the impact of near capacity operations on customers, evaluate system performance, and understand passenger behavior when choosing alternative routes. PIIM uses data from Automatic Fare Collection (AFC) and Automatic Vehicle Location (AVL) systems and is applicable to general urban rail networks with entry and exit fare transactions. PIIM consists of three main modules: the left behind model, the route fractions inference model, and the inference model. The left behind model estimates the probability of any passenger being left behind by station and time interval. The route fractions inference model estimates the route choice fractions given the left behind probabilities. The inference model maps each pair of fare transaction records (entry and exit) to a set of feasible itineraries and infers the train(s) boarded by passengers based on the left behind probabilities and the route choice fractions. Synthetic data using AFC and AVL data from a major, congested urban rail system were generated to validate the model. The results show that the model accurately predicts the itineraries used by passengers, the left behind probabilities, route choice fractions, train loads, and other performance metrics of interest. PIIM is also used with actual data to illustrate its applicability.