Travel Demand and Traffic Prediction with Cell Phone Data: Calibration by Mathematical Program with Equilibrium Constraints

Abstract

Transportation models allow for prediction of travel demands and design of interventions to improve the network performance. An essential component of such models is the origin-destination matrix, which is traditionally generated using roadside and/or household surveys. These surveys are expensive, time consuming and do not capture temporal variation in travel demand. Anonymised location data from cell phones present an alternative source of mobility information which is passively collected, widely available and naturally captures temporal trends. However, these data contain other biases which must be corrected for using more reliable data. In this study, data from the Radio Network Controller of the Andorran telecom company is combined with limited traffic count data in order to develop a calibrated urban transportation model. An initial trip matrix is generated from the telecom data and a parameterized correction model is used to modify the trip matrix before predicting traffic. The parameters of the correction model are optimized by solving a Mathematical Program with Equilibrium Constraints. Outof-sample predictions from the calibrated model are shown to agree well with actual traffic volumes. This approach can reduce or eliminate the need for travel surveys while improving understanding of travel demands and traffic.