Travel time prediction based on historical trajectory data

Abstract

Travel time prediction could be applied to various fields and purposes. For traffic managers, travel time prediction is a fundamental part of traffic system operation. Its results may assist traffic management department in adjusting traffic flow through time-dependent rules. From the travellers' viewpoints, travel time prediction saves travel time and improves reliability through the selection of travel routes pre-trip and en route to optimize travel plans. A large number of research efforts on travel time prediction have been conducted, but trip travel time prediction is relatively limited, compared with link travel time. Travellers are more interested in specific trip travel time than average link travel time. The advances in positioning technologies, such as Global Positioning System (GPS), make it possible to collect a large number of vehicle trajectories which cover the whole road network as long as data are enough and is growing as an alternative data set for travel time prediction as well as other traffic studies. In view of these, we extend the conventional methodology of link travel time prediction to trip using historical trajectory data from taxis in urban road network. This article basically consists of the following several parts, extracting origins and destinations, searching for matched trips, testing for normal distribution, detecting and removing outliers, predicting travel time in a statistic way, and evaluating the reliability of prediction results. Experiments based on taxi data in Shenzhen are conducted and the results are evaluated.