Estimation and prediction of dynamic matrix travel on a public transport corridor using historical data and real-time information

Abstract

In this paper a new methodology to estimate/update and forecast dynamic real time origin–destination travel matrices (OD) for a public transport corridor is presented. The main objective is to use available historical data, and combine it with online information regarding the entry and exit of each particular user (e.g. through the fare system, FS), to make predictions and updates for the OD matrices. The proposed methodology consists of two parts: (1) an estimation algorithm for OD matrices of public transport (EODPT), and (2) a prediction algorithm (PODPT) based on artificial neural networks (ANNs). The EODPT is based on a model that incorporates the travel time distribution between OD pairs and the modeling of the travel destination choice as a multinomial distribution, which is updated using a Bayesian approach with new available information. This approach makes it possible to correct the estimates of both the current OD interval matrices and of preceding intervals. The proposed approach was tested using actual demand data for the Metro of Valparaiso corridor in Chile (Merval), and simulated travel information in the corridor. The results are compared favorably with a static approach and can support the use of this methodology in real applications. The execution times obtained in the test cases do not exceed 10 s.