Transit Vehicle Arrival Prediction: Algorithm and Large-Scale Implementation

Abstract

An algorithm is presented to predict transit vehicle arrival times up to 1 h in advance. It uses the time series of data from an automated vehicle location system, consisting of time and location pairs. These data are used with historical statistics in an optimal filtering framework to predict future arrivals. The algorithm is implemented for a large transit fleet in Seattle, Washington, and the prediction results for hundreds of locations are made widely available on the Web. An evaluation of the second busiest but most complex prediction site is presented to demonstrate the value of prediction over the use of schedules alone.