Probabilistic Urban Link Travel Time Estimation Model Using Large-Scale Taxi Trip Data

Xianyuan Zhan,Satish V. Ukkusuri
2015-01-01
Abstract:Accurate estimation and prediction of urban link travel times are critical for urban traffic operations and management. This paper develops a probabilistic mixture model to estimate urban link travel times using large-scale taxi trip data. Unlike typical Global Positioning System (GPS) trace data, the taxi trip data used in this study provide limited trip level information, which contain only the locations of origin and destination, travel times and distances, etc. The focus of this study is to develop a robust probabilistic link travel time estimation model and demonstrate the feasibility of estimating network conditions from large-scale origin-destination (OD) level trip data. In the model, the path taken by a taxi is considered as latent and taxi drivers’ route choice behaviors are modeled using a multinomial logit distribution. The observed path data given the possible paths set and the link travel time parameters can be thus characterized using a mixture distribution. An efficient solution approach based on expectation-optimization (EM) algorithm is proposed to solve the problem. The model is tested on estimating the distribution of the link travel times for 30 min time periods using real-world data from Midtown Manhattan, New York City. Robust estimation results are obtained owing to the adoption of a probabilistic framework.
What problem does this paper attempt to address?