Short-term estimation for transit vehicle locations using AVL information and historical data

Tracking transit vehicles is helpful for users to know precisely where their next ride is. However, due to the low updates in the real-time vehicle position services, tracking how the vehicle moves each second is not always possible. In this work, the NextLocation algorithm is presented for estimating the vehicle's location each second using the last vehicle's real-time speed and location and the historical speed and location of vehicles in the surrounding area and time of the last vehicle's real-time location. The algorithm takes the real-time location and speed of the vehicle and selects the vehicles close to it (in time and distance) in the historical data set, then averages their historical instant speeds. The average speed is used for estimating the vehicle's location in the next second. The NextLocation algorithm estimates the vehicle location in milliseconds, since it only needs to perform geographical selections and basic computations, making it ideal for real-life scenarios. A study case is presented in a data set containing 6,900 locations, and discussions about the algorithm's accuracy are presented. The best algorithm's accuracy is obtained when the distance between two consecutive real-time vehicle locations is below 60 meters with a time difference below 20 seconds. The algorithm's accuracy decays when consecutive real-time locations are further apart in time and distance. The NextLocation can be used on a web map for live tracking vehicle locations and also could be used by transit operators for monitoring purposes. However, NextLocation does not consider the accuracy errors generated by the GPS devices installed in the vehicles and does not consider reroutes of breakdowns which might limit the usability for transit operators. How to overcome this limitation will be studied in future research.