Improving the travel time prediction accuracy for trucks based on historical GPS data

Improving the travel time prediction accuracy for trucks based on historical GPS data

Weerheim, Ruben (TU Delft Mechanical, Maritime and Materials Engineering)

Jiang, X. (mentor)
Negenborn, R.R. (graduation committee)

Delft University of Technology

Marine Technology | Transport Engineering and Logistics

2019-11-27

Route planning is an important part for companies that transport goods between different locations. To optimize the route planning, it is important that the travel time predictions come close to reality. A too tight schedule would lead to pressure to the driver, late fees and reputation damage, but a too loose schedule would result in wasted capacity. To predict the travel time, first a speed has to be assigned to each road in the digital map, after which the shortest path can be calculated. This research proposes a speed prediction model based on neural networks to predict the speed of each road such that the travel time prediction accuracy is improved. The neural network speed prediction model will be compared to the current speed prediction model, which is based on random forest. The speed prediction model first learns from preprocessed GPS data that is obtained from two different companies that operate with trucks. After relationships have been found between the driven speed and road properties (e.g. speed limit, road width, tunnel, etc.), the speed prediction model predicts a speed for each road in the digital map. Subsequently, the shortest path can be calculated from the first to the last GPS point of different routes. Then the predicted travel times from the shortest paths were compared to the real driven travel times. After comparison, it turned out that for both GPS data sets, the neural network speed prediction model did not outperform the travel time prediction accuracy in sMdAPE_TT of the random forest speed prediction model.

Travel Time Prediction Accuracy
GPS Data
Truck
Neural Network

http://resolver.tudelft.nl/uuid:083695ca-a32a-4699-a6f1-55f06edaf5f3

Student theses

master thesis

© 2019 Ruben Weerheim